Concept graph

The terms of the Study Guide and the links between them

The 450 terms of the index of terms, each linked to the terms it is studied with. Most links were written by hand, from what the essays say about one another. The rest come from the study pages: two terms introduced in the same section, or named together in a recall question and its answer. A dashed line only joins consecutive terms of an essay that would otherwise fall apart.

Click a concept to bring it to the center. The concepts linked to it form the ring around it; the rest keep their direction and shrink toward the rim with distance. The color of a concept is its track.

Back to the Study Guide · Index of terms

How a model keeps learning

How Can a Reward Train a Neural Network?

policy

Introduced in How Can a Reward Train a Neural Network? · study page · How a model keeps learning

Linked to: Markov state; policy iteration; bootstrapping; on-policy and off-policy; amortization (How Does a Policy Improve?); exponential tilt (How Does a Policy Improve?); teacher and support (Who Teaches the Weights?); GQ(λ) (Learning About What You Do Not Do); importance ratio (Learning About What You Do Not Do); rollout and policy (Filling the Stores).

Markov state

Introduced in How Can a Reward Train a Neural Network? · study page · How a model keeps learning

Linked to: policy; macrostate as predictor (The Demon Is a Learner); belief state (Which World Model?).

return and discount factor

Introduced in How Can a Reward Train a Neural Network? · study page · How a model keeps learning

Linked to: λ-return and eligibility trace; general value function (Learning About What You Do Not Do).

state value and action value

Introduced in How Can a Reward Train a Neural Network? · study page · How a model keeps learning

Linked to: Bellman expectation equation; policy iteration; Bellman optimality equations; baseline and advantage; general value function (Learning About What You Do Not Do); successor features (Learning About What You Do Not Do); momentum as the price of a constraint (The Standard of Motion).

Bellman expectation equation

Introduced in How Can a Reward Train a Neural Network? · study page · How a model keeps learning

Linked to: state value and action value; policy-improvement argument (How Does a Policy Improve?); projected Bellman error and gradient TD (Learning About What You Do Not Do).

Bellman optimality equations

Introduced in How Can a Reward Train a Neural Network? · study page · How a model keeps learning

Linked to: state value and action value; policy iteration; flow conservation (How Does a Policy Improve?).

temporal-difference error

Introduced in How Can a Reward Train a Neural Network? · study page · How a model keeps learning

Linked to: bootstrapping; λ-return and eligibility trace; on-policy and off-policy; GQ(λ) (Learning About What You Do Not Do); delta-rule memory (Where Experience Goes).

on-policy and off-policy

Introduced in How Can a Reward Train a Neural Network? · study page · How a model keeps learning

Linked to: policy; temporal-difference error; bootstrapping; λ-return and eligibility trace; teacher and support (Who Teaches the Weights?); deadly triad (Learning About What You Do Not Do); projected Bellman error and gradient TD (Learning About What You Do Not Do); stale rollout (Filling the Stores); five remedies for stale keys (How LLMs Could Place Explicit Memories); bandit and exploration (The Bookmaker Is a Learner).

REINFORCE

Introduced in How Can a Reward Train a Neural Network? · study page · How a model keeps learning

Linked to: GRPO; reparameterization trick (How Does a Policy Improve?).

baseline and advantage

Introduced in How Can a Reward Train a Neural Network? · study page · How a model keeps learning

Linked to: state value and action value; GRPO; advantage-weighted regression (How Does a Policy Improve?).

generalized advantage estimation

Introduced in How Can a Reward Train a Neural Network? · study page · How a model keeps learning

Linked to: λ-return and eligibility trace; GRPO; GQ(λ) (Learning About What You Do Not Do).

GRPO

Introduced in How Can a Reward Train a Neural Network? · study page · How a model keeps learning

Linked to: REINFORCE; baseline and advantage; generalized advantage estimation; forgetting within a task (Who Teaches the Weights?); Vector Policy Optimization (Who Teaches the Weights?); importance ratio (Learning About What You Do Not Do); rollout and policy (Filling the Stores); stale rollout (Filling the Stores); reward hacking (Filling the Stores).

How Does a Policy Improve?

policy-improvement argument

Introduced in How Does a Policy Improve? · study page · How a model keeps learning

Linked to: discounted occupancy; Bellman expectation equation (How Can a Reward Train a Neural Network?); policy iteration (How Can a Reward Train a Neural Network?); projected Bellman error and gradient TD (Learning About What You Do Not Do); generalized policy improvement (Learning About What You Do Not Do).

three jobs of a world model

Introduced in How Does a Policy Improve? · study page · How a model keeps learning

Linked to: discounted occupancy; amortization; reparameterization trick; model-predictive control; successor features (Learning About What You Do Not Do); transition model (Which World Model?); five capacities of a world model (Which World Model?); JEPA (Which World Model?).

amortization

Introduced in How Does a Policy Improve? · study page · How a model keeps learning

Linked to: three jobs of a world model; reparameterization trick; model-predictive control; policy (How Can a Reward Train a Neural Network?); amortization argument for weights (Where Experience Goes); rollout and policy (Filling the Stores); self-distillation on a run’s own rollouts (Filling the Stores); stale keys (How LLMs Could Place Explicit Memories); amortization as learning and as efficiency (The Given and the Found); equilibrium internalization (The Given and the Found).

reparameterization trick

Introduced in How Does a Policy Improve? · study page · How a model keeps learning

Linked to: three jobs of a world model; amortization; model-predictive control; REINFORCE (How Can a Reward Train a Neural Network?).

mass-covering and mode-seeking KL

Introduced in How Does a Policy Improve? · study page · How a model keeps learning

Linked to: exponential tilt; temperature as a Lagrange multiplier; self-distillation with privileged context (Who Teaches the Weights?); coverage condition (Learning About What You Do Not Do); consolidation as an M-step (Where Experience Goes); fluctuation theorem and Jarzynski equality (The Demon Is a Learner: The Mathematics).

exponential tilt

Introduced in How Does a Policy Improve? · study page · How a model keeps learning

Linked to: mass-covering and mode-seeking KL; temperature as a Lagrange multiplier; search as regularized policy improvement; four questions for reading an algorithm; policy (How Can a Reward Train a Neural Network?); self-distillation with privileged context (Who Teaches the Weights?); implicit value of a context (Who Teaches the Weights?); mirror descent (The Bookmaker Is a Learner); cost-function market and the LMSR (The Bookmaker Is a Learner); Gibbs state (The Demon Is a Learner: The Mathematics).

temperature as a Lagrange multiplier

Introduced in How Does a Policy Improve? · study page · How a model keeps learning

Linked to: mass-covering and mode-seeking KL; exponential tilt; control as inference; forgetting within a task (Who Teaches the Weights?); liquidity parameter (The Bookmaker Is a Learner); shadow price (The Bookmaker Is a Learner: The Mathematics); temperature as an exchange rate (The Demon Is a Learner: The Mathematics); Gibbs state (The Demon Is a Learner: The Mathematics); momentum as the price of a constraint (The Standard of Motion).

E-step and M-step of policy improvement

Introduced in How Does a Policy Improve? · study page · How a model keeps learning

Linked to: control as inference; restricted ELBO; improvement separated from fitting; consolidation as an M-step (Where Experience Goes); free energy (The Demon Is a Learner: The Mathematics).

flow conservation

Introduced in How Does a Policy Improve? · study page · How a model keeps learning

Linked to: discounted occupancy; Bellman optimality equations (How Can a Reward Train a Neural Network?); shadow price (The Bookmaker Is a Learner: The Mathematics).

improvement separated from fitting

Introduced in How Does a Policy Improve? · study page · How a model keeps learning

Linked to: E-step and M-step of policy improvement; self-distillation with privileged context (Who Teaches the Weights?).

advantage-weighted regression

Introduced in How Does a Policy Improve? · study page · How a model keeps learning

Linked to: return-conditioned policy; baseline and advantage (How Can a Reward Train a Neural Network?).

aleatoric and epistemic uncertainty

Introduced in How Does a Policy Improve? · study page · How a model keeps learning

Linked to: return-conditioned policy.

search as regularized policy improvement

Introduced in How Does a Policy Improve? · study page · How a model keeps learning

Linked to: model-predictive control; exponential tilt; Direct Preference Optimization; bandit and exploration (The Bookmaker Is a Learner); manufactured target (The Given and the Found).

Direct Preference Optimization

Introduced in How Does a Policy Improve? · study page · How a model keeps learning

Linked to: search as regularized policy improvement; implicit value of a context (Who Teaches the Weights?); moving teacher and fixed anchor (Who Teaches the Weights?).

four questions for reading an algorithm

Introduced in How Does a Policy Improve? · study page · How a model keeps learning

Linked to: exponential tilt; consolidation as an M-step (Where Experience Goes).

Who Teaches the Weights?

teacher and support

Introduced in Who Teaches the Weights? · study page · How a model keeps learning

Linked to: implicit value of a context; usable information; moving teacher and fixed anchor; policy (How Can a Reward Train a Neural Network?); on-policy and off-policy (How Can a Reward Train a Neural Network?); distilling a selector from the attention it replaces (Filling the Stores); on-policy distillation of specialists (Filling the Stores).

self-distillation with privileged context

Introduced in Who Teaches the Weights? · study page · How a model keeps learning

Linked to: implicit value of a context; usable information; mass-covering and mode-seeking KL (How Does a Policy Improve?); exponential tilt (How Does a Policy Improve?); improvement separated from fitting (How Does a Policy Improve?); consolidation as an M-step (Where Experience Goes); self-distillation on a run’s own rollouts (Filling the Stores); logical omniscience (Before the Search Settles); grounded and ungrounded targets (The Given and the Found); zone of proximal development (Building Minds).

forgetting within a task

Introduced in Who Teaches the Weights? · study page · How a model keeps learning

Linked to: implicit value of a context; Vector Policy Optimization; GRPO (How Can a Reward Train a Neural Network?); temperature as a Lagrange multiplier (How Does a Policy Improve?); coverage condition (Learning About What You Do Not Do); self-distillation on a run’s own rollouts (Filling the Stores); bandit and exploration (The Bookmaker Is a Learner); Hedge as a bookmaker (The Bookmaker Is a Learner); Landauer’s principle (The Demon Is a Learner); depth and breadth (The Given and the Found).

Vector Policy Optimization

Introduced in Who Teaches the Weights? · study page · How a model keeps learning

Linked to: forgetting within a task; two conditions on a teacher; GRPO (How Can a Reward Train a Neural Network?); set-max policy (Learning About What You Do Not Do); per-answer weighting (Learning About What You Do Not Do); depth and breadth (The Given and the Found).

two conditions on a teacher

It must know more than the student, and its targets must be ones the student can absorb.

Introduced in Who Teaches the Weights? · study page · How a model keeps learning

Linked to: Vector Policy Optimization; on-policy distillation of specialists (Filling the Stores); Latent Sparse Attention (How LLMs Could Place Explicit Memories); zone of proximal development (Building Minds).

usable information

Introduced in Who Teaches the Weights? · study page · How a model keeps learning

Linked to: teacher and support; self-distillation with privileged context; implicit value of a context; amortization argument for weights (Where Experience Goes); Levin search (The Demon Is a Learner); H-theorem for a Markov chain (The Demon Is a Learner: The Mathematics); logical omniscience (Before the Search Settles); recoverability and learned operation (Where Does Seeing Become Thinking?).

moving teacher and fixed anchor

Introduced in Who Teaches the Weights? · study page · How a model keeps learning

Linked to: teacher and support; Direct Preference Optimization (How Does a Policy Improve?); distilling a selector from the attention it replaces (Filling the Stores); liquidity parameter (The Bookmaker Is a Learner); grounded and ungrounded targets (The Given and the Found).

Learning About What You Do Not Do

general value function

Introduced in Learning About What You Do Not Do · study page · How a model keeps learning

Linked to: deadly triad; successor features; generalized policy improvement; set-max policy; return and discount factor (How Can a Reward Train a Neural Network?); state value and action value (How Can a Reward Train a Neural Network?); ladder of stores (Where Experience Goes); correctness condition (The Given and the Found); schema mechanism (Building Minds).

deadly triad

Introduced in Learning About What You Do Not Do · study page · How a model keeps learning

Linked to: general value function; projected Bellman error and gradient TD; GQ(λ); importance ratio; bootstrapping (How Can a Reward Train a Neural Network?); on-policy and off-policy (How Can a Reward Train a Neural Network?).

projected Bellman error and gradient TD

Introduced in Learning About What You Do Not Do · study page · How a model keeps learning

Linked to: deadly triad; GQ(λ); importance ratio; Bellman expectation equation (How Can a Reward Train a Neural Network?); on-policy and off-policy (How Can a Reward Train a Neural Network?); policy-improvement argument (How Does a Policy Improve?).

GQ(λ)

Introduced in Learning About What You Do Not Do · study page · How a model keeps learning

Linked to: deadly triad; projected Bellman error and gradient TD; importance ratio; coverage condition; policy (How Can a Reward Train a Neural Network?); temporal-difference error (How Can a Reward Train a Neural Network?); λ-return and eligibility trace (How Can a Reward Train a Neural Network?); generalized advantage estimation (How Can a Reward Train a Neural Network?); schema mechanism (Building Minds).

importance ratio

Introduced in Learning About What You Do Not Do · study page · How a model keeps learning

Linked to: deadly triad; projected Bellman error and gradient TD; GQ(λ); policy (How Can a Reward Train a Neural Network?); GRPO (How Can a Reward Train a Neural Network?); rollout and policy (Filling the Stores); stale rollout (Filling the Stores); stale keys (How LLMs Could Place Explicit Memories); bandit and exploration (The Bookmaker Is a Learner); fluctuation theorem and Jarzynski equality (The Demon Is a Learner: The Mathematics).

generalized policy improvement

Introduced in Learning About What You Do Not Do · study page · How a model keeps learning

Linked to: general value function; successor features; set-max policy; policy iteration (How Can a Reward Train a Neural Network?); policy-improvement argument (How Does a Policy Improve?); return-conditioned policy (How Does a Policy Improve?).

per-answer weighting

Introduced in Learning About What You Do Not Do · study page · How a model keeps learning

Linked to: successor features; Vector Policy Optimization (Who Teaches the Weights?).

coverage condition

Introduced in Learning About What You Do Not Do · study page · How a model keeps learning

Linked to: GQ(λ); mass-covering and mode-seeking KL (How Does a Policy Improve?); forgetting within a task (Who Teaches the Weights?); bandit and exploration (The Bookmaker Is a Learner).

Where Experience Goes

learned update rule

Introduced in Where Experience Goes · study page · How a model keeps learning

Linked to: delta-rule memory; Gated DeltaNet and Kimi Delta Attention (Every Layer a Store); underdetermination of the landscape (The Given and the Found).

ladder of stores

Introduced in Where Experience Goes · study page · How a model keeps learning

Linked to: delta-rule memory; general value function (Learning About What You Do Not Do); four questions of a store (Every Layer a Store); stores written in training and stores written while reading (Every Layer a Store); Memorilla (How LLMs Could Place Explicit Memories); three placements of a memory (How LLMs Could Place Explicit Memories); dreaming phase (Which AGI?); Complementary Learning Systems (Cognitive Architectures); catastrophic interference (Cognitive Architectures).

delta-rule memory

Introduced in Where Experience Goes · study page · How a model keeps learning

Linked to: learned update rule; ladder of stores; consolidation as an M-step; temporal-difference error (How Can a Reward Train a Neural Network?); outer-product store (Every Layer a Store); interference between stored keys (Every Layer a Store); Gated DeltaNet and Kimi Delta Attention (Every Layer a Store); catastrophic interference (Cognitive Architectures).

consolidation as an M-step

Introduced in Where Experience Goes · study page · How a model keeps learning

Linked to: delta-rule memory; amortization argument for weights; mass-covering and mode-seeking KL (How Does a Policy Improve?); E-step and M-step of policy improvement (How Does a Policy Improve?); four questions for reading an algorithm (How Does a Policy Improve?); self-distillation with privileged context (Who Teaches the Weights?); on-policy distillation of specialists (Filling the Stores); deflationary and substantive answers about consolidation (How LLMs Could Place Explicit Memories).

amortization argument for weights

Introduced in Where Experience Goes · study page · How a model keeps learning

Linked to: consolidation as an M-step; amortization (How Does a Policy Improve?); usable information (Who Teaches the Weights?); break-even for consolidation (How LLMs Could Place Explicit Memories); deflationary and substantive answers about consolidation (How LLMs Could Place Explicit Memories); compiled/interpreted gap (Which AGI?); library learning (What Should a Program Inherit?); amortization as learning and as efficiency (The Given and the Found).

Every Layer a Store

mixture of experts

Introduced in Every Layer a Store · study page · How a model keeps learning

Linked to: four questions of a store.

four questions of a store

What writes the store, when, how much it holds before its entries interfere, and what reading it costs.

Introduced in Every Layer a Store · study page · How a model keeps learning

Linked to: mixture of experts; outer-product store; ladder of stores (Where Experience Goes).

outer-product store

Introduced in Every Layer a Store · study page · How a model keeps learning

Linked to: four questions of a store; interference between stored keys; delta-rule memory (Where Experience Goes); tensor product representation (The Algebraic Mind Meets the Neural World).

interference between stored keys

Introduced in Every Layer a Store · study page · How a model keeps learning

Linked to: outer-product store; stores written in training and stores written while reading; delta-rule memory (Where Experience Goes); context extension (Filling the Stores); distraction issue (How LLMs Could Place Explicit Memories); mean field and the cavity method (The Demon Is a Learner: The Mathematics); tensor product representation (The Algebraic Mind Meets the Neural World); catastrophic interference (Cognitive Architectures); superposition (Dispersed Minds, Simulated Selves).

stores written in training and stores written while reading

The feed-forward layer holds what the model knows; the attention cache holds what the model has been told.

Introduced in Every Layer a Store · study page · How a model keeps learning

Linked to: interference between stored keys; clean recall and picking the right answer; ladder of stores (Where Experience Goes); plateau before the facts (Filling the Stores); explicit memory (How LLMs Could Place Explicit Memories); lexicon-grammar continuum as feature continuum (A Parser Without a Grammar); operator/operand split (A Parser Without a Grammar); gating (Dispersed Minds, Simulated Selves).

five ways to keep the past of a context

Full history, a rolling window, selected entries, compressed entries, a recurrent state.

Introduced in Every Layer a Store · study page · How a model keeps learning

Linked to: clean recall and picking the right answer; selector.

selector

A second, cheaper store in front of the cache; it scores keys and the main attention reads only the winners.

Introduced in Every Layer a Store · study page · How a model keeps learning

Linked to: five ways to keep the past of a context; Gated DeltaNet and Kimi Delta Attention; pretraining, mid-training, SFT, RL, distillation and drafter training (Filling the Stores); distilling a selector from the attention it replaces (Filling the Stores); dense warm-up (Filling the Stores); Sinkhorn iteration (Filling the Stores); open process and open weights (Filling the Stores); Latent Sparse Attention (How LLMs Could Place Explicit Memories).

Gated DeltaNet and Kimi Delta Attention

Introduced in Every Layer a Store · study page · How a model keeps learning

Linked to: selector; store over depth; learned update rule (Where Experience Goes); delta-rule memory (Where Experience Goes).

store over depth

A memory written and read by the layers of one token’s pass.

Introduced in Every Layer a Store · study page · How a model keeps learning

Linked to: Gated DeltaNet and Kimi Delta Attention; Manifold-Constrained Hyper-Connections; monotone merge (Merge and Selection); persistent carrier (Experience, Selfhood, and the Organization of a Mind); residual stream as blackboard (Dispersed Minds, Simulated Selves).

Manifold-Constrained Hyper-Connections

Introduced in Every Layer a Store · study page · How a model keeps learning

Linked to: store over depth; Attention Residuals; Sinkhorn iteration (Filling the Stores).

five meanings of depth

Introduced in Every Layer a Store · study page · How a model keeps learning

Linked to: Attention Residuals; expressive power and learnability (The Algebraic Mind Meets the Neural World); depth and breadth (The Given and the Found); outer loop (Dispersed Minds, Simulated Selves).

Filling the Stores

pretraining, mid-training, SFT, RL, distillation and drafter training

Introduced in Filling the Stores · study page · How a model keeps learning

Linked to: rollout and policy; distilling a selector from the attention it replaces; dense warm-up; reward hacking; open process and open weights; selector (Every Layer a Store); Systems A, B and M (Which World Model?).

rollout and policy

Introduced in Filling the Stores · study page · How a model keeps learning

Linked to: pretraining, mid-training, SFT, RL, distillation and drafter training; plateau before the facts; stale rollout; reward hacking; self-distillation on a run’s own rollouts; policy (How Can a Reward Train a Neural Network?); GRPO (How Can a Reward Train a Neural Network?); amortization (How Does a Policy Improve?); importance ratio (Learning About What You Do Not Do); stale keys (How LLMs Could Place Explicit Memories).

plateau before the facts

The stretch of training in which a model knows the kind of answer a template wants and not which answer.

Introduced in Filling the Stores · study page · How a model keeps learning

Linked to: rollout and policy; Muon; stores written in training and stores written while reading (Every Layer a Store).

distilling a selector from the attention it replaces

Introduced in Filling the Stores · study page · How a model keeps learning

Linked to: pretraining, mid-training, SFT, RL, distillation and drafter training; dense warm-up; teacher and support (Who Teaches the Weights?); moving teacher and fixed anchor (Who Teaches the Weights?); selector (Every Layer a Store); Latent Sparse Attention (How LLMs Could Place Explicit Memories).

Sinkhorn iteration

Introduced in Filling the Stores · study page · How a model keeps learning

Linked to: Muon; context extension; selector (Every Layer a Store); Manifold-Constrained Hyper-Connections (Every Layer a Store); combinatorial wall (The Bookmaker Is a Learner).

context extension

Introduced in Filling the Stores · study page · How a model keeps learning

Linked to: Sinkhorn iteration; interference between stored keys (Every Layer a Store).

stale rollout

Introduced in Filling the Stores · study page · How a model keeps learning

Linked to: rollout and policy; reward hacking; on-policy and off-policy (How Can a Reward Train a Neural Network?); GRPO (How Can a Reward Train a Neural Network?); importance ratio (Learning About What You Do Not Do); stale keys (How LLMs Could Place Explicit Memories).

on-policy distillation of specialists

Introduced in Filling the Stores · study page · How a model keeps learning

Linked to: reward hacking; teacher and support (Who Teaches the Weights?); two conditions on a teacher (Who Teaches the Weights?); consolidation as an M-step (Where Experience Goes).

self-distillation on a run’s own rollouts

SFT on rollouts collected from an RL run, started from a mid-trained checkpoint, after which RL resumes.

Introduced in Filling the Stores · study page · How a model keeps learning

Linked to: rollout and policy; amortization (How Does a Policy Improve?); self-distillation with privileged context (Who Teaches the Weights?); forgetting within a task (Who Teaches the Weights?); stale keys (How LLMs Could Place Explicit Memories).

open process and open weights

Introduced in Filling the Stores · study page · How a model keeps learning

Linked to: pretraining, mid-training, SFT, RL, distillation and drafter training; selector (Every Layer a Store).

How LLMs Could Place Explicit Memories

explicit memory

A fact the weights do not contain, found where it was written down.

Introduced in How LLMs Could Place Explicit Memories · study page · How a model keeps learning

Linked to: stale keys; stores written in training and stores written while reading (Every Layer a Store).

stale keys

Keys read by newer weights than the ones that wrote them.

Introduced in How LLMs Could Place Explicit Memories · study page · How a model keeps learning

Linked to: explicit memory; five remedies for stale keys; amortization (How Does a Policy Improve?); importance ratio (Learning About What You Do Not Do); rollout and policy (Filling the Stores); stale rollout (Filling the Stores); self-distillation on a run’s own rollouts (Filling the Stores); persistent inquiry with replaceable interpreters (What Persists When the Agents Change?).

distraction issue

Introduced in How LLMs Could Place Explicit Memories · study page · How a model keeps learning

Linked to: Memorilla; interference between stored keys (Every Layer a Store).

Latent Sparse Attention

Introduced in How LLMs Could Place Explicit Memories · study page · How a model keeps learning

Linked to: Memorilla; hippocampal indexing; two conditions on a teacher (Who Teaches the Weights?); selector (Every Layer a Store); distilling a selector from the attention it replaces (Filling the Stores).

deflationary and substantive answers about consolidation

Consolidation as amortization, against consolidation as representation learning.

Introduced in How LLMs Could Place Explicit Memories · study page · How a model keeps learning

Linked to: break-even for consolidation; consolidation as an M-step (Where Experience Goes); amortization argument for weights (Where Experience Goes); compiled/interpreted gap (Which AGI?); representational conjecture (What Persists When the Agents Change?); amortization as learning and as efficiency (The Given and the Found).

three placements of a memory

By position, where it arrived; by content, what it resembles; into connections, what it taught.

Introduced in How LLMs Could Place Explicit Memories · study page · How a model keeps learning

Linked to: hippocampal indexing; ladder of stores (Where Experience Goes); Complementary Learning Systems (Cognitive Architectures).

five remedies for stale keys

Introduced in How LLMs Could Place Explicit Memories · study page · How a model keeps learning

Linked to: stale keys; Memorilla; on-policy and off-policy (How Can a Reward Train a Neural Network?).

Learners, bets and heat

The Bookmaker Is a Learner

the relay

The essay’s name for Hanson’s market scoring rule: a trader is paid the score of their forecast minus the score of the forecast they replaced.

Introduced in The Bookmaker Is a Learner · study page · Learners, bets and heat

Linked to: liquidity parameter; constant-function market maker; proper scoring rule; market scoring rule; cost-function market and the LMSR.

liquidity parameter

The b of the LMSR, which bounds the subsidy at b\log n and acts as an inverse learning rate.

Introduced in The Bookmaker Is a Learner · study page · Learners, bets and heat

Linked to: the relay; regret and no-regret learning; bandit and exploration; mirror descent; cost-function market and the LMSR; Hedge as a bookmaker; follow the regularized leader; temperature as a Lagrange multiplier (How Does a Policy Improve?); moving teacher and fixed anchor (Who Teaches the Weights?); temperature as an exchange rate (The Demon Is a Learner: The Mathematics).

bandit and exploration

Introduced in The Bookmaker Is a Learner · study page · Learners, bets and heat

Linked to: liquidity parameter; regret and no-regret learning; mirror descent; Hedge as a bookmaker; on-policy and off-policy (How Can a Reward Train a Neural Network?); search as regularized policy improvement (How Does a Policy Improve?); forgetting within a task (Who Teaches the Weights?); importance ratio (Learning About What You Do Not Do); coverage condition (Learning About What You Do Not Do); non-dogmatism (Before the Search Settles).

mirror descent

Introduced in The Bookmaker Is a Learner · study page · Learners, bets and heat

Linked to: liquidity parameter; regret and no-regret learning; bandit and exploration; Hedge as a bookmaker; follow the regularized leader; exponential tilt (How Does a Policy Improve?); overdamped limit (The Demon Is a Learner); Polyak–Łojasiewicz condition (The Demon Is a Learner); thermodynamic length (The Demon Is a Learner: The Mathematics).

matching as pricing at zero temperature

As b\to0 the LMSR cost function tends to the house’s worst-case liability, which is what matching computes.

Introduced in The Bookmaker Is a Learner · study page · Learners, bets and heat

Linked to: constant-function market maker; marginal polytope; subset betting and pair betting; combinatorial wall; abnormality with a price (Presumed Normal).

marginal polytope

Introduced in The Bookmaker Is a Learner · study page · Learners, bets and heat

Linked to: matching as pricing at zero temperature; subset betting and pair betting; combinatorial wall; mean field and the cavity method (The Demon Is a Learner: The Mathematics); multicalibration (Before the Search Settles).

proper scoring rule

Introduced in The Bookmaker Is a Learner · study page · Learners, bets and heat

Linked to: the relay; constant-function market maker; market scoring rule; sequential convex pari-mutuel mechanism (The Bookmaker Is a Learner: The Mathematics); conjugate representation of a cost function (The Bookmaker Is a Learner: The Mathematics); table of adversaries (Before the Search Settles).

market scoring rule

Introduced in The Bookmaker Is a Learner · study page · Learners, bets and heat

Linked to: the relay; proper scoring rule; cost-function market and the LMSR; sequential convex pari-mutuel mechanism (The Bookmaker Is a Learner: The Mathematics); conjugate representation of a cost function (The Bookmaker Is a Learner: The Mathematics).

cost-function market and the LMSR

Introduced in The Bookmaker Is a Learner · study page · Learners, bets and heat

Linked to: the relay; liquidity parameter; constant-function market maker; market scoring rule; follow the regularized leader; exponential tilt (How Does a Policy Improve?); sequential convex pari-mutuel mechanism (The Bookmaker Is a Learner: The Mathematics); conjugate representation of a cost function (The Bookmaker Is a Learner: The Mathematics); Gibbs state (The Demon Is a Learner: The Mathematics); logical induction criterion (Before the Search Settles).

combinatorial wall

Over the n! finishing orders of a race, matching pair bets is NP-hard and exact LMSR pricing is #P-hard for both betting languages.

Introduced in The Bookmaker Is a Learner · study page · Learners, bets and heat

Linked to: matching as pricing at zero temperature; marginal polytope; subset betting and pair betting; Sinkhorn iteration (Filling the Stores); logical induction criterion (Before the Search Settles); presumption of independence (Before the Search Settles).

wealth of traders as a second learner

Across many questions a market reweights its traders through their accumulated money, multiplicatively for those who stake a fixed fraction of their wealth, as Hedge reweights experts.

Introduced in The Bookmaker Is a Learner · study page · Learners, bets and heat

Linked to: Hedge as a bookmaker; nonequilibrium free energy (The Demon Is a Learner: The Mathematics); Solomonoff’s prior (Before the Search Settles); logical induction criterion (Before the Search Settles); process, deployment and lineage (Which AGI?); five conditions of mentality (Minds by Degrees); residual integration (Is the Monastery Minded?).

The Bookmaker Is a Learner: The Mathematics

sequential convex pari-mutuel mechanism

Introduced in The Bookmaker Is a Learner: The Mathematics · study page · Learners, bets and heat

Linked to: shadow price; proper scoring rule (The Bookmaker Is a Learner); market scoring rule (The Bookmaker Is a Learner); cost-function market and the LMSR (The Bookmaker Is a Learner).

conjugate representation of a cost function

Introduced in The Bookmaker Is a Learner: The Mathematics · study page · Learners, bets and heat

Linked to: shadow price; proper scoring rule (The Bookmaker Is a Learner); market scoring rule (The Bookmaker Is a Learner); cost-function market and the LMSR (The Bookmaker Is a Learner); follow the regularized leader (The Bookmaker Is a Learner); free energy (The Demon Is a Learner: The Mathematics).

Presumed Normal

adaptive logic

A lower limit logic, a set of abnormalities presumed false, and a strategy for reading the premises as normally as possible.

Introduced in Presumed Normal · study page · Learners, bets and heat

Linked to: Dab-formula and unreliable set; Reliability and Minimal Abnormality; condition and marking; final derivability; rebutting and undercutting defeaters; nogood; Self and idealized sources (An Ontology That Pays for Itself); outcome indistinguishability (Before the Search Settles); gaps and gluts (Before the Search Settles); joint constraint abduction (What Must Be True?).

Dab-formula and unreliable set

Introduced in Presumed Normal · study page · Learners, bets and heat

Linked to: adaptive logic; Reliability and Minimal Abnormality; nogood.

internal and external dynamics

Introduced in Presumed Normal · study page · Learners, bets and heat

Linked to: condition and marking; final derivability; logical omniscience (Before the Search Settles).

rebutting and undercutting defeaters

Introduced in Presumed Normal · study page · Learners, bets and heat

Linked to: adaptive logic; Reliability and Minimal Abnormality; argumentation framework and extension; presumption of independence (Before the Search Settles); reason, price and defeater (Before the Search Settles).

here-and-there and equilibrium model

Introduced in Presumed Normal · study page · Learners, bets and heat

Linked to: Reliability and Minimal Abnormality; stable model and reduct.

abnormality with a price

A weight w on each abnormality, which turns the selection of minimal models into a distribution over models and plays the part of an inverse temperature.

Introduced in Presumed Normal · study page · Learners, bets and heat

Linked to: Reliability and Minimal Abnormality; condition and marking; nogood; matching as pricing at zero temperature (The Bookmaker Is a Learner); presumption of independence (Before the Search Settles); gaps and gluts (Before the Search Settles).

condition as an abductive answer

The condition of an adaptive derivation of G, read as a hypothesis of normality under which the premises entail G.

Introduced in Presumed Normal · study page · Learners, bets and heat

Linked to: condition and marking; deduction, generalization and abduction (What Must Be True?); joint constraint abduction (What Must Be True?).

An Ontology That Pays for Itself

two-part code

Introduced in An Ontology That Pays for Itself · study page · Learners, bets and heat

Linked to: seed and stalk; best-parse code; Kolmogorov complexity (The Demon Is a Learner); physical entropy (The Demon Is a Learner); reason, price and defeater (Before the Search Settles); library learning (What Should a Program Inherit?).

best-parse code

Coding a sentence by its single best parse and then the words given the parse, at a cost of -\log P(t^*\mid x) bits above the best the grammar allows.

Introduced in An Ontology That Pays for Itself · study page · Learners, bets and heat

Linked to: seed and stalk; two-part code; three-part code and realization channel.

three-part code and realization channel

Ontology, facts given the ontology, and text given the facts and a lexicon; the last part is the channel, which should know syntax, word choice and paraphrase and nothing about the world.

Introduced in An Ontology That Pays for Itself · study page · Learners, bets and heat

Linked to: best-parse code; reporting policy.

reporting policy

The probability that a source mentions a fact, given the fact, the source and the audience it writes for.

Introduced in An Ontology That Pays for Itself · study page · Learners, bets and heat

Linked to: three-part code and realization channel; common ground; provenance judgment.

role tuple

\langle A,R,S,B\rangle\equiv\langle B,S,R,A\rangle: A plays role R toward B, which plays role S.

Introduced in An Ontology That Pays for Itself · study page · Learners, bets and heat

Linked to: seed and stalk; two-way reading; image; distributive IS-A.

two-way reading

An asserted class-level tuple claims that every A stands in the relation to some B and every B stands in it to some A.

Introduced in An Ontology That Pays for Itself · study page · Learners, bets and heat

Linked to: role tuple; image; distributive IS-A.

distributive IS-A

IS-A says only what holds of every member, so a fact about a kind needs a singular term for the kind.

Introduced in An Ontology That Pays for Itself · study page · Learners, bets and heat

Linked to: role tuple; two-way reading; image.

provenance judgment

\mathrm{Prov}(T,P), source P asserts T: the only form of judgment, never itself a term.

Introduced in An Ontology That Pays for Itself · study page · Learners, bets and heat

Linked to: reporting policy; Self and idealized sources; table of adversaries (Before the Search Settles).

Self and idealized sources

The system as the last judge, and contexts such as a novel’s canon treated as sources; only these are closed under inference.

Introduced in An Ontology That Pays for Itself · study page · Learners, bets and heat

Linked to: provenance judgment; adaptive logic (Presumed Normal); gaps and gluts (Before the Search Settles).

The Demon Is a Learner

overdamped limit

Introduced in The Demon Is a Learner · study page · Learners, bets and heat

Linked to: Polyak–Łojasiewicz condition; three geometries of gradient flow; mirror descent (The Bookmaker Is a Learner); fluctuation–dissipation relation (The Demon Is a Learner: The Mathematics); Aristotle’s first and second law (The Standard of Motion); structural stability (The Standard of Motion); Hamiltonian vector field (The Standard of Motion: The Mathematics).

Polyak–Łojasiewicz condition

Introduced in The Demon Is a Learner · study page · Learners, bets and heat

Linked to: overdamped limit; three geometries of gradient flow; mirror descent (The Bookmaker Is a Learner); log-Sobolev inequality (The Demon Is a Learner: The Mathematics); Morse–Smale flow (The Standard of Motion: The Mathematics).

three geometries of gradient flow

Weights under constant drag, densities under the Wasserstein distance, and probabilities under the Fisher drag; in each, velocity follows force.

Introduced in The Demon Is a Learner · study page · Learners, bets and heat

Linked to: overdamped limit; Polyak–Łojasiewicz condition; Boltzmann’s assumption; free energy (The Demon Is a Learner: The Mathematics); thermodynamic length (The Demon Is a Learner: The Mathematics); Aristotle’s first and second law (The Standard of Motion).

Boltzmann’s assumption

Two balls about to collide are independent.

Introduced in The Demon Is a Learner · study page · Learners, bets and heat

Linked to: three geometries of gradient flow; coarse graining; Kolmogorov complexity; score; relevant and irrelevant memory; Stosszahlansatz (The Demon Is a Learner: The Mathematics); mean field and the cavity method (The Demon Is a Learner: The Mathematics); presumption of independence (Before the Search Settles); JEPA (Which World Model?); Aristotle’s first and second law (The Standard of Motion); Liouville’s theorem (The Standard of Motion).

coarse graining

Introduced in The Demon Is a Learner · study page · Learners, bets and heat

Linked to: Boltzmann’s assumption; Kolmogorov complexity; score; macrostate and microstate (The Demon Is a Learner: The Mathematics); total correlation (The Demon Is a Learner: The Mathematics); JEPA (Which World Model?); physics as a coarse-graining (Presence and Actuality); bounded and persistent observer (Quantum Mechanics, Observers, and the Ruliad); Aristotle’s first and second law (The Standard of Motion); Liouville’s theorem (The Standard of Motion).

Landauer’s principle

Introduced in The Demon Is a Learner · study page · Learners, bets and heat

Linked to: physical entropy; score; relevant and irrelevant memory; forgetting within a task (Who Teaches the Weights?); entropy production (The Demon Is a Learner: The Mathematics); nonequilibrium free energy (The Demon Is a Learner: The Mathematics); Aristotle’s first and second law (The Standard of Motion); Liouville’s theorem (The Standard of Motion).

Kolmogorov complexity

Introduced in The Demon Is a Learner · study page · Learners, bets and heat

Linked to: Boltzmann’s assumption; coarse graining; physical entropy; Levin search; two-part code (An Ontology That Pays for Itself); Solomonoff’s prior (Before the Search Settles); JEPA (Which World Model?); Aristotle’s first and second law (The Standard of Motion); Liouville’s theorem (The Standard of Motion).

physical entropy

Zurek’s sum of the missing information and the algorithmic information content of the record.

Introduced in The Demon Is a Learner · study page · Learners, bets and heat

Linked to: Landauer’s principle; Kolmogorov complexity; two-part code (An Ontology That Pays for Itself).

score

Introduced in The Demon Is a Learner · study page · Learners, bets and heat

Linked to: Boltzmann’s assumption; coarse graining; Landauer’s principle; relevant and irrelevant memory; H-theorem for a Markov chain (The Demon Is a Learner: The Mathematics); entropy production (The Demon Is a Learner: The Mathematics); detailed balance (The Demon Is a Learner: The Mathematics).

relevant and irrelevant memory

Introduced in The Demon Is a Learner · study page · Learners, bets and heat

Linked to: Boltzmann’s assumption; Landauer’s principle; score; macrostate as predictor; nonequilibrium free energy (The Demon Is a Learner: The Mathematics); information bottleneck (The Demon Is a Learner: The Mathematics).

macrostate as predictor

A state that predicts its own future; what a predictor needs to keep.

Introduced in The Demon Is a Learner · study page · Learners, bets and heat

Linked to: relevant and irrelevant memory; Levin search; Markov state (How Can a Reward Train a Neural Network?); macrostate and microstate (The Demon Is a Learner: The Mathematics); reason, price and defeater (Before the Search Settles); belief state (Which World Model?); physics as a coarse-graining (Presence and Actuality); bounded and persistent observer (Quantum Mechanics, Observers, and the Ruliad).

The Demon Is a Learner: The Mathematics

macrostate and microstate

Introduced in The Demon Is a Learner: The Mathematics · study page · Learners, bets and heat

Linked to: temperature as an exchange rate; coarse graining (The Demon Is a Learner); macrostate as predictor (The Demon Is a Learner).

temperature as an exchange rate

The energy that buys one nat of entropy.

Introduced in The Demon Is a Learner: The Mathematics · study page · Learners, bets and heat

Linked to: macrostate and microstate; Gibbs state; temperature as a Lagrange multiplier (How Does a Policy Improve?); liquidity parameter (The Bookmaker Is a Learner); shadow price (The Bookmaker Is a Learner: The Mathematics).

total correlation

Introduced in The Demon Is a Learner: The Mathematics · study page · Learners, bets and heat

Linked to: Stosszahlansatz; coarse graining (The Demon Is a Learner); Liouville’s theorem (The Standard of Motion).

H-theorem for a Markov chain

Introduced in The Demon Is a Learner: The Mathematics · study page · Learners, bets and heat

Linked to: Stosszahlansatz; usable information (Who Teaches the Weights?); score (The Demon Is a Learner).

fluctuation–dissipation relation

Introduced in The Demon Is a Learner: The Mathematics · study page · Learners, bets and heat

Linked to: Gibbs state; overdamped limit (The Demon Is a Learner).

entropy production

Introduced in The Demon Is a Learner: The Mathematics · study page · Learners, bets and heat

Linked to: log-Sobolev inequality; detailed balance; Landauer’s principle (The Demon Is a Learner); score (The Demon Is a Learner).

nonequilibrium free energy

Introduced in The Demon Is a Learner: The Mathematics · study page · Learners, bets and heat

Linked to: log-Sobolev inequality; information bottleneck; wealth of traders as a second learner (The Bookmaker Is a Learner); Landauer’s principle (The Demon Is a Learner); relevant and irrelevant memory (The Demon Is a Learner).

information bottleneck

Introduced in The Demon Is a Learner: The Mathematics · study page · Learners, bets and heat

Linked to: nonequilibrium free energy; relevant and irrelevant memory (The Demon Is a Learner).

fluctuation theorem and Jarzynski equality

Introduced in The Demon Is a Learner: The Mathematics · study page · Learners, bets and heat

Linked to: free energy; detailed balance; thermodynamic length; mean field and the cavity method; mass-covering and mode-seeking KL (How Does a Policy Improve?); importance ratio (Learning About What You Do Not Do).

mean field and the cavity method

Introduced in The Demon Is a Learner: The Mathematics · study page · Learners, bets and heat

Linked to: free energy; detailed balance; fluctuation theorem and Jarzynski equality; thermodynamic length; interference between stored keys (Every Layer a Store); marginal polytope (The Bookmaker Is a Learner); Boltzmann’s assumption (The Demon Is a Learner).

Before the Search Settles

logical omniscience

Introduced in Before the Search Settles · study page · Learners, bets and heat

Linked to: Solomonoff’s prior; self-distillation with privileged context (Who Teaches the Weights?); usable information (Who Teaches the Weights?); internal and external dynamics (Presumed Normal); bounded rationality (Cognitive Architectures).

logical prior

Demski’s distribution over theories: draw sentences at random and keep each one that does not contradict those already kept.

Introduced in Before the Search Settles · study page · Learners, bets and heat

Linked to: Solomonoff’s prior; Gaifman inductivity; semidecision procedure (What Must Be True?).

Solomonoff’s prior

Introduced in Before the Search Settles · study page · Learners, bets and heat

Linked to: logical omniscience; logical prior; Gaifman inductivity; regret and no-regret learning (The Bookmaker Is a Learner); Hedge as a bookmaker (The Bookmaker Is a Learner); wealth of traders as a second learner (The Bookmaker Is a Learner); Kolmogorov complexity (The Demon Is a Learner); Levin search (The Demon Is a Learner).

logical induction criterion

Introduced in Before the Search Settles · study page · Learners, bets and heat

Linked to: non-dogmatism; defensive forecasting; No Expected Net Update; regret and no-regret learning (The Bookmaker Is a Learner); cost-function market and the LMSR (The Bookmaker Is a Learner); combinatorial wall (The Bookmaker Is a Learner); wealth of traders as a second learner (The Bookmaker Is a Learner); amortizing search versus amortizing discovery (The Given and the Found).

non-dogmatism

Introduced in Before the Search Settles · study page · Learners, bets and heat

Linked to: Gaifman inductivity; logical induction criterion; bandit and exploration (The Bookmaker Is a Learner).

paradox resistance

Introduced in Before the Search Settles · study page · Learners, bets and heat

Linked to: defensive forecasting; multicalibration; condition and marking (Presumed Normal); stable model and reduct (Presumed Normal).

outcome indistinguishability

Introduced in Before the Search Settles · study page · Learners, bets and heat

Linked to: defensive forecasting; multicalibration; table of adversaries; adaptive logic (Presumed Normal).

table of adversaries

Where the adversary sits, inside the believer or outside, against what holds it to account, settlement over time or checks on demand.

Introduced in Before the Search Settles · study page · Learners, bets and heat

Linked to: defensive forecasting; multicalibration; outcome indistinguishability; presumption of independence; independent check; proper scoring rule (The Bookmaker Is a Learner); provenance judgment (An Ontology That Pays for Itself).

presumption of independence

Introduced in Before the Search Settles · study page · Learners, bets and heat

Linked to: table of adversaries; reason, price and defeater; combinatorial wall (The Bookmaker Is a Learner); rebutting and undercutting defeaters (Presumed Normal); abnormality with a price (Presumed Normal); Boltzmann’s assumption (The Demon Is a Learner).

reason, price and defeater

A reason is a latent structure with independence presumed beyond it; a price is the marginal probability it implies for one sentence; a defeater is structure found outside it.

Introduced in Before the Search Settles · study page · Learners, bets and heat

Linked to: presumption of independence; gaps and gluts; rebutting and undercutting defeaters (Presumed Normal); two-part code (An Ontology That Pays for Itself); macrostate as predictor (The Demon Is a Learner).

gaps and gluts

Introduced in Before the Search Settles · study page · Learners, bets and heat

Linked to: reason, price and defeater; adaptive logic (Presumed Normal); Reliability and Minimal Abnormality (Presumed Normal); abnormality with a price (Presumed Normal); Self and idealized sources (An Ontology That Pays for Itself).

independent check

A check that does not depend on the beliefs being trained, whether it is internal or external.

Introduced in Before the Search Settles · study page · Learners, bets and heat

Linked to: table of adversaries; convergence as a learned proxy (The Given and the Found); amplitude and sign of a learning signal (The Given and the Found); grounded and ungrounded targets (The Given and the Found).

No Expected Net Update

Introduced in Before the Search Settles · study page · Learners, bets and heat

Linked to: logical induction criterion.

Which intelligence

Poetry as a Mode of Thinking

organ of perception

Hirshfield’s thesis that a poem discovers what is imperceptible without it.

Introduced in Poetry as a Mode of Thinking · study page · Which intelligence

Linked to: generative constraint; operations of discovery (The Operations of Discovery).

generative constraint

A formal restriction, such as sound-patterning, that makes otherwise unlikely associations discoverable.

Introduced in Poetry as a Mode of Thinking · study page · Which intelligence

Linked to: organ of perception; forced adjacency; register collision.

renewable surprise

Hirshfield’s term for discovery that cannot be cached and must be re-enacted at each reading.

Introduced in Poetry as a Mode of Thinking · study page · Which intelligence

Linked to: fake thinking; precision through compression.

fake thinking

Carlsmith’s term for the form of cognition without its function.

Introduced in Poetry as a Mode of Thinking · study page · Which intelligence

Linked to: renewable surprise; attunement; meaning without the mouth; slop (The Operations of Discovery).

attunement

Carlsmith’s term for meaning-laden receptivity to the world.

Introduced in Poetry as a Mode of Thinking · study page · Which intelligence

Linked to: fake thinking; evidential weight (The Operations of Discovery); temporal coherence (The Operations of Discovery).

forced adjacency

Placing together things that analytic thinking must keep separate.

Introduced in Poetry as a Mode of Thinking · study page · Which intelligence

Linked to: generative constraint; thinking in negative space; precision through compression; register collision; operations of discovery (The Operations of Discovery); evidential weight (The Operations of Discovery); slop (The Operations of Discovery).

thinking in negative space

Letting what is not said carry the content.

Introduced in Poetry as a Mode of Thinking · study page · Which intelligence

Linked to: forced adjacency; precision through compression; register collision; operations of discovery (The Operations of Discovery); evidential weight (The Operations of Discovery); slop (The Operations of Discovery).

precision through compression

A density that would become less precise if unpacked.

Introduced in Poetry as a Mode of Thinking · study page · Which intelligence

Linked to: renewable surprise; forced adjacency; thinking in negative space; register collision; operations of discovery (The Operations of Discovery); slop (The Operations of Discovery).

register collision

Unlike vocabularies meeting in a small textual space.

Introduced in Poetry as a Mode of Thinking · study page · Which intelligence

Linked to: generative constraint; forced adjacency; thinking in negative space; precision through compression; operations of discovery (The Operations of Discovery); evidential weight (The Operations of Discovery); slop (The Operations of Discovery); Neptune-proof (The Operations of Discovery).

meaning without the mouth

An LLM’s access to what a poem says without its sound, rhythm and breath as vehicle.

Introduced in Poetry as a Mode of Thinking · study page · Which intelligence

Linked to: fake thinking; vehicle and content of a concept (What Is a Concept?); meta-epistemic grounding (What Is a Concept?).

The Operations of Discovery

breadth and depth

Tao’s distinction between what AI tools and human researchers each excel at.

Introduced in The Operations of Discovery · study page · Which intelligence

Linked to: jumping machines; trajectory and front (Which AGI?); depth and breadth (The Given and the Found).

jumping machines

Tao’s image for tools that solve a problem in one leap or fail, with no partial progress.

Introduced in The Operations of Discovery · study page · Which intelligence

Linked to: breadth and depth; jaggedness; temporal coherence; persistent inquiry with replaceable interpreters (What Persists When the Agents Change?).

operations of discovery

The four operations of poetic thinking, claimed to be those of discovery in general.

Introduced in The Operations of Discovery · study page · Which intelligence

Linked to: evidential weight; slop; Neptune-proof; organ of perception (Poetry as a Mode of Thinking); forced adjacency (Poetry as a Mode of Thinking); thinking in negative space (Poetry as a Mode of Thinking); precision through compression (Poetry as a Mode of Thinking); register collision (Poetry as a Mode of Thinking).

evidential weight

The pressure of a real unsolved problem, brought by someone for whom it is unresolved.

Introduced in The Operations of Discovery · study page · Which intelligence

Linked to: operations of discovery; slop; jaggedness; temporal coherence; Neptune-proof; attunement (Poetry as a Mode of Thinking); forced adjacency (Poetry as a Mode of Thinking); thinking in negative space (Poetry as a Mode of Thinking); register collision (Poetry as a Mode of Thinking).

slop

In the deeper sense, output that is coherent and formally sophisticated but shaped by pressures that do not track truth.

Introduced in The Operations of Discovery · study page · Which intelligence

Linked to: operations of discovery; evidential weight; Neptune-proof; fake thinking (Poetry as a Mode of Thinking); forced adjacency (Poetry as a Mode of Thinking); thinking in negative space (Poetry as a Mode of Thinking); precision through compression (Poetry as a Mode of Thinking); register collision (Poetry as a Mode of Thinking).

jaggedness

Karpathy’s word for a model that is brilliant and childish at once.

Introduced in The Operations of Discovery · study page · Which intelligence

Linked to: jumping machines; evidential weight; temporal coherence; efficacy-matching (Which AGI?); motley mix (Dispersed Minds, Simulated Selves).

temporal coherence

One observer’s sustained encounter with a problem over years, which accumulates orientation.

Introduced in The Operations of Discovery · study page · Which intelligence

Linked to: jumping machines; evidential weight; jaggedness; attunement (Poetry as a Mode of Thinking); compiled/interpreted gap (Which AGI?); trajectory and front (Which AGI?).

Neptune-proof

Said of a result that the next data point will not destroy, as Neptune destroyed Bode’s law.

Introduced in The Operations of Discovery · study page · Which intelligence

Linked to: operations of discovery; evidential weight; slop; register collision (Poetry as a Mode of Thinking).

Which AGI?

process, deployment and lineage

The three referents of “AGI”: a network in its inference loop, the agent built around it, and the succession of such agents that a lab, its users and its retraining pipeline constitute.

Introduced in Which AGI? · study page · Which intelligence

Linked to: the ladder; cog view; self-sustainment; wealth of traders as a second learner (The Bookmaker Is a Learner); Systems A, B and M (Which World Model?); epistemic action (Where Does Seeing Become Thinking?).

the ladder

Six dimensions along which a system was successively required to be unrestricted: domain, modality, horizon, memory, autonomy, ends.

Introduced in Which AGI? · study page · Which intelligence

Linked to: process, deployment and lineage; maximal pole; graded mentality (Minds by Degrees).

maximal pole

The position that the ladder has no natural break between intelligence and agency.

Introduced in Which AGI? · study page · Which intelligence

Linked to: the ladder.

cog view

General intelligence as coverage of a set of functional roles, each of which may be absent, underpowered or filled outside the network.

Introduced in Which AGI? · study page · Which intelligence

Linked to: process, deployment and lineage; profile; efficacy-matching; self-sustainment; five conditions of mentality (Minds by Degrees); cognitive architecture (Cognitive Architectures).

profile

An assignment of fillers to the functional roles.

Introduced in Which AGI? · study page · Which intelligence

Linked to: cog view; efficacy-matching; dreaming phase; compiled/interpreted gap.

efficacy-matching

Two profiles attain comparable general capability without being behaviorally equivalent.

Introduced in Which AGI? · study page · Which intelligence

Linked to: cog view; profile; jaggedness (The Operations of Discovery); internality (Minds by Degrees); iconic and discursive formats (What Is a Concept?).

dreaming phase

A consolidation pass in which the same model re-reads, merges and prunes its memory files after activity.

Introduced in Which AGI? · study page · Which intelligence

Linked to: profile; compiled/interpreted gap; ladder of stores (Where Experience Goes); cognitive cycle (Cognitive Architectures); Complementary Learning Systems (Cognitive Architectures).

compiled/interpreted gap

A skill followed as written instructions against a skill consolidated into fluent, implicit, composable competence.

Introduced in Which AGI? · study page · Which intelligence

Linked to: profile; dreaming phase; trajectory and front; amortization argument for weights (Where Experience Goes); deflationary and substantive answers about consolidation (How LLMs Could Place Explicit Memories); temporal coherence (The Operations of Discovery); Systems A, B and M (Which World Model?); persistent inquiry with replaceable interpreters (What Persists When the Agents Change?); chunking (Cognitive Architectures).

trajectory and front

A single agent’s cumulative, path-dependent learning against what the lineage as a whole has absorbed.

Introduced in Which AGI? · study page · Which intelligence

Linked to: compiled/interpreted gap; breadth and depth (The Operations of Discovery); temporal coherence (The Operations of Discovery); model identity and trajectory identity (What Persists When the Agents Change?); representational conjecture (What Persists When the Agents Change?); secondhand inheritance (Building Minds).

self-sustainment

The lineage produces the data, research and capital for its own next generation.

Introduced in Which AGI? · study page · Which intelligence

Linked to: process, deployment and lineage; cog view; Systems A, B and M (Which World Model?).

What Must Be True?

deduction, generalization and abduction

Peirce’s three operations as the thesis uses them: work out consequences, find a common description of cases, supply the missing condition.

Introduced in What Must Be True? · study page · Which intelligence

Linked to: precondition and postcondition; joint constraint abduction; anti-unification; condition as an abductive answer (Presumed Normal); amortizing search versus amortizing discovery (The Given and the Found).

precondition and postcondition

Introduced in What Must Be True? · study page · Which intelligence

Linked to: deduction, generalization and abduction; algorithm C (What Should a Program Inherit?).

joint constraint abduction

Find one condition A with D_i\land A\models C_i for every case i, each D_i\land A consistent.

Introduced in What Must Be True? · study page · Which intelligence

Linked to: deduction, generalization and abduction; semidecision procedure; anti-unification; adaptive logic (Presumed Normal); Reliability and Minimal Abnormality (Presumed Normal); condition and marking (Presumed Normal); final derivability (Presumed Normal); condition as an abductive answer (Presumed Normal).

semidecision procedure

Introduced in What Must Be True? · study page · Which intelligence

Linked to: joint constraint abduction; constant-test fragment; final derivability (Presumed Normal); Levin search (The Demon Is a Learner); logical prior (Before the Search Settles).

constant-test fragment

Premises only test variables against the two constants, and conclusions contain no constructors.

Introduced in What Must Be True? · study page · Which intelligence

Linked to: semidecision procedure.

anti-unification

Introduced in What Must Be True? · study page · Which intelligence

Linked to: deduction, generalization and abduction; joint constraint abduction; three achievements of an open-ended reasoner; linear schema (What Should a Program Inherit?); library learning (What Should a Program Inherit?); e-graph (What Should a Program Inherit?).

three achievements of an open-ended reasoner

Generating admissible hypotheses, selecting and testing them under a budget, revising the framing.

Introduced in What Must Be True? · study page · Which intelligence

Linked to: anti-unification; revisable abstraction (Where Does Seeing Become Thinking?).

Which World Model?

level and trajectory claims

“this system falls short” reports a level of performance; “this approach is a dead end” is a claim about a trend, and needs a metric, its movement across systems, and an expectation.

Introduced in Which World Model? · study page · Which intelligence

Linked to: possessing versus using a representation.

possessing versus using a representation

A representation is a model when its organization supports inferences about the represented system, and explanatory when the agent uses those inferences; competence shows neither where the structure sits nor how it is used.

Introduced in Which World Model? · study page · Which intelligence

Linked to: level and trajectory claims; transition model; hosting and judging perspectives (What Persists When the Agents Change?); three levels of evidence (Where Does Seeing Become Thinking?); recoverability and learned operation (Where Does Seeing Become Thinking?); three tiers of understanding (Dispersed Minds, Simulated Selves).

transition model

The control sense of “world model”: a predictor of the next state from an estimated state and a candidate action.

Introduced in Which World Model? · study page · Which intelligence

Linked to: possessing versus using a representation; five capacities of a world model; three jobs of a world model (How Does a Policy Improve?); efference copy (Where Does Seeing Become Thinking?).

five capacities of a world model

Representing structure, tracking the current situation, predicting consequences, using a model, and maintaining a model; they vary independently.

Introduced in Which World Model? · study page · Which intelligence

Linked to: transition model; goal compass; three jobs of a world model (How Does a Policy Improve?); source, format and operations (Where Does Seeing Become Thinking?).

goal compass

Beckmann, Queloz and Freitas’s name for a directional representation, separate from the map, that steers travel toward a distant goal.

Introduced in Which World Model? · study page · Which intelligence

Linked to: five capacities of a world model; belief state.

belief state

In the NextLat paper’s sense: a hidden state sufficient for predicting the future sequence.

Introduced in Which World Model? · study page · Which intelligence

Linked to: goal compass; JEPA; Markov state (How Can a Reward Train a Neural Network?); macrostate as predictor (The Demon Is a Learner); synthetic item (Building Minds).

JEPA

Joint Embedding Predictive Architecture: predict representations of targets and not the targets themselves.

Introduced in Which World Model? · study page · Which intelligence

Linked to: belief state; Systems A, B and M; three jobs of a world model (How Does a Policy Improve?); Boltzmann’s assumption (The Demon Is a Learner); coarse graining (The Demon Is a Learner); Kolmogorov complexity (The Demon Is a Learner); Levin search (The Demon Is a Learner); revisable abstraction (Where Does Seeing Become Thinking?); Aristotle’s first and second law (The Standard of Motion).

Systems A, B and M

Dupoux, LeCun and Malik’s division: learning from observation, learning through action, and the control of their interaction.

Introduced in Which World Model? · study page · Which intelligence

Linked to: JEPA; cost of elicitation; pretraining, mid-training, SFT, RL, distillation and drafter training (Filling the Stores); process, deployment and lineage (Which AGI?); compiled/interpreted gap (Which AGI?); self-sustainment (Which AGI?).

cost of elicitation

The examples, additional training, search, tool calls and experimenter-supplied information needed to bring a capacity out.

Introduced in Which World Model? · study page · Which intelligence

Linked to: Systems A, B and M; recoverability and learned operation (Where Does Seeing Become Thinking?).

What Persists When the Agents Change?

model identity and trajectory identity

Copies of one model given one prompt still make different attempts, and each attempt becomes part of its own later context.

Introduced in What Persists When the Agents Change? · study page · Which intelligence

Linked to: persistent inquiry with replaceable interpreters; trajectory and front (Which AGI?).

persistent inquiry with replaceable interpreters

Wójcik’s proposal: explicit records carry an inquiry’s history so that it is recoverable across interpreters.

Introduced in What Persists When the Agents Change? · study page · Which intelligence

Linked to: model identity and trajectory identity; transactive memory; stale keys (How LLMs Could Place Explicit Memories); jumping machines (The Operations of Discovery); compiled/interpreted gap (Which AGI?); provenance graph (What Should a Program Inherit?).

transactive memory

A shared repository of trajectories from which other agents retrieve, matched to their current state.

Introduced in What Persists When the Agents Change? · study page · Which intelligence

Linked to: persistent inquiry with replaceable interpreters; local memory.

local memory

Explicit, role-specific state that gives a shared conversation a different retained interpretation for each participant.

Introduced in What Persists When the Agents Change? · study page · Which intelligence

Linked to: transactive memory; hosting and judging perspectives; representational conjecture.

hosting and judging perspectives

Representing two traditions in shared machinery against assessing their claims well.

Introduced in What Persists When the Agents Change? · study page · Which intelligence

Linked to: local memory; possessing versus using a representation (Which World Model?); primary persona (Dispersed Minds, Simulated Selves).

representational conjecture

That training insulated within a trajectory reaches representational spaces which shared training and contextual adaptation do not.

Introduced in What Persists When the Agents Change? · study page · Which intelligence

Linked to: local memory; deflationary and substantive answers about consolidation (How LLMs Could Place Explicit Memories); trajectory and front (Which AGI?).

What Should a Program Inherit?

genealogical identity

A module stays a descendant of the same module while its implementation changes; duplication creates a new identity.

Introduced in What Should a Program Inherit? · study page · Which intelligence

Linked to: linear schema; provenance graph.

linear schema

A common pattern in which each schema variable occurs only once, so that every hole can vary independently.

Introduced in What Should a Program Inherit? · study page · Which intelligence

Linked to: genealogical identity; library learning; anti-unification (What Must Be True?).

library learning

Finding definitions that let a collection of programs be rewritten more economically, paying for the definitions as well as their uses.

Introduced in What Should a Program Inherit? · study page · Which intelligence

Linked to: linear schema; e-graph; amortization argument for weights (Where Experience Goes); two-part code (An Ontology That Pays for Itself); anti-unification (What Must Be True?); chunking (Cognitive Architectures).

e-graph

Introduced in What Should a Program Inherit? · study page · Which intelligence

Linked to: library learning; algorithm C; anti-unification (What Must Be True?).

algorithm C

The 2005 thesis’s procedure that constructs a term from a sequence of choices alongside type inference, returning a term and a substitution.

Introduced in What Should a Program Inherit? · study page · Which intelligence

Linked to: e-graph; precondition and postcondition (What Must Be True?).

provenance graph

A record of which versions, documents and evaluation results were supplied to an attempt and which edits it produced.

Introduced in What Should a Program Inherit? · study page · Which intelligence

Linked to: genealogical identity; persistent inquiry with replaceable interpreters (What Persists When the Agents Change?).

Where Does Seeing Become Thinking?

source, format and operations

Where information comes from, how it is represented internally, and which operations are available over the representation; none determines the others.

Introduced in Where Does Seeing Become Thinking? · study page · Which intelligence

Linked to: iconic representation; five capacities of a world model (Which World Model?); content, format, origin and dynamics (What Is a Concept?); format turn (What Is a Concept?).

iconic representation

Block’s sense, stated more exactly than in the concept survey: analog tracking and mirroring, in which degrees of difference in what is represented correspond to degrees of difference in the representation.

Introduced in Where Does Seeing Become Thinking? · study page · Which intelligence

Linked to: source, format and operations; three levels of evidence; content-tracking difference (What Phenomenal Intuition Gets Right); iconic and discursive formats (What Is a Concept?); perceptual pluralism (What Is a Concept?).

three levels of evidence

Recovering a variable from internal states, showing by intervention that it affects the answer, and explaining the transformation the model performs.

Introduced in Where Does Seeing Become Thinking? · study page · Which intelligence

Linked to: iconic representation; recoverability and learned operation; possessing versus using a representation (Which World Model?); Language of Thought (What Is a Concept?).

recoverability and learned operation

Information that could in principle answer a question, against a procedure that makes it available at a particular cost and reliability.

Introduced in Where Does Seeing Become Thinking? · study page · Which intelligence

Linked to: three levels of evidence; efference copy; usable information (Who Teaches the Weights?); possessing versus using a representation (Which World Model?); cost of elicitation (Which World Model?).

epistemic action

Kirsh and Maglio’s term for a move that makes information easier to obtain or compute, as against one that directly advances the objective.

Introduced in Where Does Seeing Become Thinking? · study page · Which intelligence

Linked to: efference copy; revisable abstraction; process, deployment and lineage (Which AGI?); meta-epistemic grounding (What Is a Concept?).

revisable abstraction

Keeping access to the source and revising an abstraction when the task changes.

Introduced in Where Does Seeing Become Thinking? · study page · Which intelligence

Linked to: epistemic action; three achievements of an open-ended reasoner (What Must Be True?); JEPA (Which World Model?).

Grammar and search inside networks

The Algebraic Mind Meets the Neural World

symbol manipulation (Marcus’s three hypotheses)

Introduced in The Algebraic Mind Meets the Neural World · study page · Grammar and search inside networks

Linked to: words and rules; training independence; Language of Thought (What Is a Concept?).

tensor product representation

Introduced in The Algebraic Mind Meets the Neural World · study page · Grammar and search inside networks

Linked to: registers and treelets; Continuity and Compositionality Principles; outer-product store (Every Layer a Store); interference between stored keys (Every Layer a Store); Harmony and scheduled quantization (Merge and Selection); Language of Thought (What Is a Concept?); 2-simplicial attention (Dispersed Minds, Simulated Selves).

Continuity and Compositionality Principles

Introduced in The Algebraic Mind Meets the Neural World · study page · Grammar and search inside networks

Linked to: tensor product representation; generations of neurocompositional computing; gradient blend (Merge and Selection); Language of Thought (What Is a Concept?); symbolic data and quantitative metadata (Cognitive Architectures).

attention as approximate unbinding

Introduced in The Algebraic Mind Meets the Neural World · study page · Grammar and search inside networks

Linked to: expressive power and learnability; query as valence specification (Merge and Selection); binding problem of stacked attention (Merge and Selection); working-memory pointers (Cognitive Architectures).

Merge and Selection

monotone merge

The residual stream only adds, so it builds structure as a constraint grammar joins descriptions, not as a rewrite grammar replaces symbols.

Introduced in Merge and Selection · study page · Grammar and search inside networks

Linked to: target directs, sources supply, MLP assembles; store over depth (Every Layer a Store); Merge-without-Move (A Parser Without a Grammar); grammar as process (A Parser Without a Grammar); override as additive cancellation (A Parser Without a Grammar); parser without a grammar (A Parser Without a Grammar); residual stream as blackboard (Dispersed Minds, Simulated Selves).

target directs, sources supply, MLP assembles

The three-way split of what unification does in one structure-driven step.

Introduced in Merge and Selection · study page · Grammar and search inside networks

Linked to: monotone merge; query as valence specification; the selector is not in the value space; un-bundling (A Parser Without a Grammar).

query as valence specification

The query says what a position seeks, as a head’s valence list says what it must combine with.

Introduced in Merge and Selection · study page · Grammar and search inside networks

Linked to: target directs, sources supply, MLP assembles; Harmony and scheduled quantization; attention as approximate unbinding (The Algebraic Mind Meets the Neural World); un-bundling (A Parser Without a Grammar).

Harmony and scheduled quantization

Introduced in Merge and Selection · study page · Grammar and search inside networks

Linked to: query as valence specification; gradient blend; Gibbs state (The Demon Is a Learner: The Mathematics); tensor product representation (The Algebraic Mind Meets the Neural World); generations of neurocompositional computing (The Algebraic Mind Meets the Neural World); depth and breadth (The Given and the Found); limit cycle as terminal behavior (The Given and the Found).

gradient blend

A single continuous state superposing partial structures, not a probability distribution over parses.

Introduced in Merge and Selection · study page · Grammar and search inside networks

Linked to: Harmony and scheduled quantization; Continuity and Compositionality Principles (The Algebraic Mind Meets the Neural World); collapse of strata (A Parser Without a Grammar); grammar as process (A Parser Without a Grammar); parser without a grammar (A Parser Without a Grammar); depth and breadth (The Given and the Found).

binding problem of stacked attention

Introduced in Merge and Selection · study page · Grammar and search inside networks

Linked to: missing selection relation and bag problem; generations of neurocompositional computing (The Algebraic Mind Meets the Neural World); attention as approximate unbinding (The Algebraic Mind Meets the Neural World).

the selector is not in the value space

The query drives each combination and is never among the vectors the merge sums.

Introduced in Merge and Selection · study page · Grammar and search inside networks

Linked to: target directs, sources supply, MLP assembles; missing selection relation and bag problem.

missing selection relation and bag problem

The two manifestations of the selector’s exclusion: no record of which position governed, and no record of which daughter paired with which.

Introduced in Merge and Selection · study page · Grammar and search inside networks

Linked to: binding problem of stacked attention; the selector is not in the value space.

A Parser Without a Grammar

formalisms as coordinates

Each grammatical tradition fixes one axis of design, and the architecture’s position on each axis is read off; none is a candidate for being the Transformer’s grammar.

Introduced in A Parser Without a Grammar · study page · Grammar and search inside networks

Linked to: Merge-without-Move; un-bundling; parser without a grammar; dichotomic maps (Cognitive Architectures).

Merge-without-Move

Combination that only adds, with no operation that relocates or deletes.

Introduced in A Parser Without a Grammar · study page · Grammar and search inside networks

Linked to: formalisms as coordinates; monotone merge (Merge and Selection); Merge conjecture (Cognitive Architectures).

un-bundling

The architecture runs as separate mechanisms the design choices that a formalism’s single combinatory operation fuses.

Introduced in A Parser Without a Grammar · study page · Grammar and search inside networks

Linked to: formalisms as coordinates; collapse of strata; target directs, sources supply, MLP assembles (Merge and Selection); query as valence specification (Merge and Selection); correctness condition (The Given and the Found).

collapse of strata

Positional and functional information superimposed in one residual stream.

Introduced in A Parser Without a Grammar · study page · Grammar and search inside networks

Linked to: un-bundling; gradient blend (Merge and Selection).

lexicon-grammar continuum as feature continuum

Introduced in A Parser Without a Grammar · study page · Grammar and search inside networks

Linked to: grammar as process; operator/operand split; stores written in training and stores written while reading (Every Layer a Store); words and rules (The Algebraic Mind Meets the Neural World); superposition (Dispersed Minds, Simulated Selves).

operator/operand split

Abstractions live in the weights, which act on the activations and carry none; activation rides only token positions.

Introduced in A Parser Without a Grammar · study page · Grammar and search inside networks

Linked to: lexicon-grammar continuum as feature continuum; override as additive cancellation; stores written in training and stores written while reading (Every Layer a Store).

override as additive cancellation

Introduced in A Parser Without a Grammar · study page · Grammar and search inside networks

Linked to: operator/operand split; monotone merge (Merge and Selection).

parser without a grammar

Structure without a discrete, stored, declarative grammar object.

Introduced in A Parser Without a Grammar · study page · Grammar and search inside networks

Linked to: formalisms as coordinates; grammar as process; monotone merge (Merge and Selection); gradient blend (Merge and Selection); correctness condition (The Given and the Found); depth and breadth (The Given and the Found).

The Given and the Found

correctness condition

What makes an answer right: a fact about the task, prior to any network.

Introduced in The Given and the Found · study page · Grammar and search inside networks

Linked to: landscape and search; spectrum of givenness; amortizing search versus amortizing discovery; general value function (Learning About What You Do Not Do); un-bundling (A Parser Without a Grammar); parser without a grammar (A Parser Without a Grammar).

spectrum of givenness

How much of the correctness condition is guaranteed by construction, and how much is approximated by a surrogate trustworthy only in distribution.

Introduced in The Given and the Found · study page · Grammar and search inside networks

Linked to: correctness condition; sound but incomplete deduction.

sound but incomplete deduction

Introduced in The Given and the Found · study page · Grammar and search inside networks

Linked to: spectrum of givenness.

convergence as a learned proxy

The system’s internal signal of having settled is a trained artifact, calibrated only as far as the data that shaped it.

Introduced in The Given and the Found · study page · Grammar and search inside networks

Linked to: spurious attractor; independent check (Before the Search Settles).

depth and breadth

Iterations within one trajectory, which entrench its basin, and independent restarts, which cover more basins.

Introduced in The Given and the Found · study page · Grammar and search inside networks

Linked to: spurious attractor; limit cycle as terminal behavior; forgetting within a task (Who Teaches the Weights?); Vector Policy Optimization (Who Teaches the Weights?); five meanings of depth (Every Layer a Store); breadth and depth (The Operations of Discovery); Harmony and scheduled quantization (Merge and Selection); gradient blend (Merge and Selection); parser without a grammar (A Parser Without a Grammar).

limit cycle as terminal behavior

A single trajectory that visits the valid answers in turn and does not settle; the essay’s own suggestion.

Introduced in The Given and the Found · study page · Grammar and search inside networks

Linked to: depth and breadth; amortization as learning and as efficiency; Harmony and scheduled quantization (Merge and Selection).

amortization as learning and as efficiency

Raising the ceiling attainable at fixed data, against cutting the compute needed to reach a ceiling already attainable.

Introduced in The Given and the Found · study page · Grammar and search inside networks

Linked to: limit cycle as terminal behavior; equilibrium internalization; amortization (How Does a Policy Improve?); amortization argument for weights (Where Experience Goes); deflationary and substantive answers about consolidation (How LLMs Could Place Explicit Memories); chunking (Cognitive Architectures).

equilibrium internalization

Introduced in The Given and the Found · study page · Grammar and search inside networks

Linked to: amortization as learning and as efficiency; amortization (How Does a Policy Improve?).

manufactured target

A training target produced by a confident intermediate computation, such as a converged loop state or a search’s visit counts.

Introduced in The Given and the Found · study page · Grammar and search inside networks

Linked to: landscape and search; spurious attractor; amplitude and sign of a learning signal; grounded and ungrounded targets; underdetermination of the landscape; optimal granularity; search as regularized policy improvement (How Does a Policy Improve?).

amplitude and sign of a learning signal

The target’s confidence sets the size of the update; whether the target is grounded in truth sets its direction.

Introduced in The Given and the Found · study page · Grammar and search inside networks

Linked to: landscape and search; spurious attractor; manufactured target; grounded and ungrounded targets; underdetermination of the landscape; optimal granularity; independent check (Before the Search Settles).

underdetermination of the landscape

Supervision constrains the model only along the trajectories training visits and leaves the rest free.

Introduced in The Given and the Found · study page · Grammar and search inside networks

Linked to: landscape and search; spurious attractor; manufactured target; amplitude and sign of a learning signal; grounded and ungrounded targets; optimal granularity; learned update rule (Where Experience Goes); training independence (The Algebraic Mind Meets the Neural World).

optimal granularity

The step size at which the pull toward fewer, bigger steps balances the difficulty of learning a bigger step.

Introduced in The Given and the Found · study page · Grammar and search inside networks

Linked to: landscape and search; spurious attractor; manufactured target; amplitude and sign of a learning signal; grounded and ungrounded targets; underdetermination of the landscape.

amortizing search versus amortizing discovery

Speeding up search under a correctness condition that is given, against finding the condition.

Introduced in The Given and the Found · study page · Grammar and search inside networks

Linked to: correctness condition; Gaifman inductivity (Before the Search Settles); logical induction criterion (Before the Search Settles); deduction, generalization and abduction (What Must Be True?).

Mind and experience

Minds by Degrees

non-deflationary mentality

Real, thickly individuated structures in the system have the mental properties attributed to it; the attribution is not an interpretive gloss.

Introduced in Minds by Degrees · study page · Mind and experience

Linked to: graded mentality; ontology ladder (Can Two Minds Share One Experience); three tiers of understanding (Dispersed Minds, Simulated Selves).

graded mentality

Degrees along several partially independent dimensions, so that the useful answer is a profile.

Introduced in Minds by Degrees · study page · Mind and experience

Linked to: non-deflationary mentality; five conditions of mentality; the ladder (Which AGI?).

minds as the kind of psychology

The natural kind that a selective, structural realism about psychology’s cumulative core commits us to.

Introduced in Minds by Degrees · study page · Mind and experience

Linked to: grain-plural integration; central conjecture.

grain-plural integration

Integration without IIT’s exclusion postulate: a system can be weakly integrated at one grain and strongly at another, and both facts are real.

Introduced in Minds by Degrees · study page · Mind and experience

Linked to: minds as the kind of psychology; metastable overlap (Can Two Minds Share One Experience); three grains (The Grain of the Hard Problem); residual integration (Is the Monastery Minded?).

five conditions of mentality

Organizational unity, structured vehicles, accuracy-responsive dynamics, simulation with internal function, multi-level content.

Introduced in Minds by Degrees · study page · Mind and experience

Linked to: graded mentality; central conjecture; evaluative mentality; wealth of traders as a second learner (The Bookmaker Is a Learner); cog view (Which AGI?); ontology ladder (Can Two Minds Share One Experience); vehicle and content of a concept (What Is a Concept?).

internality

Simulation outputs function as inputs to the system’s own further processing, not only for outside consumers.

Introduced in Minds by Degrees · study page · Mind and experience

Linked to: central conjecture; reflexive closure; efficacy-matching (Which AGI?); role-process (Is the Monastery Minded?); outer loop (Dispersed Minds, Simulated Selves).

central conjecture

Psychological explanation gains valid traction on a system if and only if the five-component mechanism is running in it.

Introduced in Minds by Degrees · study page · Mind and experience

Linked to: minds as the kind of psychology; five conditions of mentality; internality.

evaluative mentality

The five conditions plus valenced states that function as evaluations within the economy.

Introduced in Minds by Degrees · study page · Mind and experience

Linked to: five conditions of mentality; reflexive closure.

reflexive closure

The economy targets its own vehicles with pointers, the targeting is a real pattern of the economy, and it is put to regulatory use.

Introduced in Minds by Degrees · study page · Mind and experience

Linked to: internality; evaluative mentality; three subjectivities; pointer content and descriptive content (Experience, Selfhood, and the Organization of a Mind); criterion of belonging (Experience, Selfhood, and the Organization of a Mind); for-me-ness, me-ness and mineness (Experience, Selfhood, and the Organization of a Mind); primary persona (Dispersed Minds, Simulated Selves).

three subjectivities

Structural perspectivality, cognitive subjectivity and agentic subjectivity, which Nagel’s formula bundles.

Introduced in Minds by Degrees · study page · Mind and experience

Linked to: reflexive closure; saturation; perspectival subjectivity (Experience, Selfhood, and the Organization of a Mind); for-me-ness, me-ness and mineness (Experience, Selfhood, and the Organization of a Mind); phenomenal-adjacent concepts (Dispersed Minds, Simulated Selves).

saturation

The mode of a vehicle whose configuration is continuously held against its target by sustained regulatory coupling over the specious present.

Introduced in Minds by Degrees · study page · Mind and experience

Linked to: three subjectivities; acquaintance (Can Two Minds Share One Experience); content-sensitive regulation (Presence and Actuality); persistent carrier (Experience, Selfhood, and the Organization of a Mind); truth-tracking and truth-bearers (Is the Monastery Minded?).

Can Two Minds Share One Experience

ontology ladder

Actual computation, minded understanding, phenomenal occurrence: three demands, each short of the next.

Introduced in Can Two Minds Share One Experience · study page · Mind and experience

Linked to: worldliness; non-deflationary mentality (Minds by Degrees); five conditions of mentality (Minds by Degrees); esse (Presence and Actuality).

worldliness

A subject and its experiences are constituted by the actual processes that sustain them, not by an abstract specification realized across worlds.

Introduced in Can Two Minds Share One Experience · study page · Mind and experience

Linked to: ontology ladder; experiential occurrence; esse (Presence and Actuality); constructive criterion (Presence and Actuality); Ruliad and rulial space (Quantum Mechanics, Observers, and the Ruliad).

experiential occurrence

A representational vehicle sustained in a particular mode of content-sensitive regulation; the proposal identifies an experience with it.

Introduced in Can Two Minds Share One Experience · study page · Mind and experience

Linked to: worldliness; acquaintance; epistemic regulation; center of epistemic gravity; shared occurrence; identity proposal (What Phenomenal Intuition Gets Right); intrinsic character (structural sense) (What Phenomenal Intuition Gets Right); availability through character (What Phenomenal Intuition Gets Right).

acquaintance

The actual regulatory involvement with representational vehicles, which requires the vehicle to exist.

Introduced in Can Two Minds Share One Experience · study page · Mind and experience

Linked to: experiential occurrence; epistemic regulation; saturation (Minds by Degrees); content-sensitive regulation (Presence and Actuality); availability through character (What Phenomenal Intuition Gets Right); presentation and uptake (What Phenomenal Intuition Gets Right); tracking, assessment and intervention (Experience, Selfhood, and the Organization of a Mind).

epistemic regulation

Regulation that addresses content through its vehicles, responsive to what they carry and how they function in the cognitive economy.

Introduced in Can Two Minds Share One Experience · study page · Mind and experience

Linked to: experiential occurrence; acquaintance; content-sensitive regulation (Presence and Actuality); tracking, assessment and intervention (Experience, Selfhood, and the Organization of a Mind).

center of epistemic gravity

The organization around which epistemic regulation coheres, named without installing an observer; adapted from Dennett’s center of narrative gravity.

Introduced in Can Two Minds Share One Experience · study page · Mind and experience

Linked to: experiential occurrence; shared occurrence; perspectival subjectivity (Experience, Selfhood, and the Organization of a Mind).

shared occurrence

One experiential occurrence jointly sustained by the compatible regulation of two distinct subjects.

Introduced in Can Two Minds Share One Experience · study page · Mind and experience

Linked to: experiential occurrence; center of epistemic gravity; metastable overlap; reality measure (Presence and Actuality); intrinsic character (structural sense) (What Phenomenal Intuition Gets Right); criterion of belonging (Experience, Selfhood, and the Organization of a Mind); field of its own (Is the Monastery Minded?); role-process (Is the Monastery Minded?).

metastable overlap

Sharing sustained over the phenomenal now by centers whose own dynamics can pull them apart.

Introduced in Can Two Minds Share One Experience · study page · Mind and experience

Linked to: shared occurrence; grain-plural integration (Minds by Degrees); field of its own (Is the Monastery Minded?); reach, latency and bandwidth (Is the Monastery Minded?).

Presence and Actuality

content-sensitive regulation

Organized activity that maintains, revises and coordinates representing states, responsive to differences relevant to their representational roles.

Introduced in Presence and Actuality · study page · Mind and experience

Linked to: vehicle role without determinate content; saturation (Minds by Degrees); acquaintance (Can Two Minds Share One Experience); epistemic regulation (Can Two Minds Share One Experience); tracking, assessment and intervention (Experience, Selfhood, and the Organization of a Mind).

vehicle role without determinate content

An occurrence keeps its place in an economy organized around representing when no particular content can be attributed to it.

Introduced in Presence and Actuality · study page · Mind and experience

Linked to: content-sensitive regulation; esse; identity proposal (What Phenomenal Intuition Gets Right); vehicle and content of a concept (What Is a Concept?).

esse

Aquinas’s act of being, borrowed by Wójcik for the actuality a theory of consciousness must account for.

Introduced in Presence and Actuality · study page · Mind and experience

Linked to: vehicle role without determinate content; operational individuation; route to determination; physics as a coarse-graining; constructive criterion; reality measure; ontology ladder (Can Two Minds Share One Experience); worldliness (Can Two Minds Share One Experience).

operational individuation

The middle step of Thomas Epistemes’s reconstruction: an objectively operative organization whose relations matter to physical interactions.

Introduced in Presence and Actuality · study page · Mind and experience

Linked to: esse; three grains (The Grain of the Hard Problem).

route to determination

What a proposed distinction needs: an account of what would make it obtain, distinguish its instances, or establish its possibility.

Introduced in Presence and Actuality · study page · Mind and experience

Linked to: esse; physics as a coarse-graining; constructive criterion; reality measure.

physics as a coarse-graining

In the Ruliad framework, bounded observers aggregate distinctions, and physical objects, states and relations emerge through that aggregation.

Introduced in Presence and Actuality · study page · Mind and experience

Linked to: esse; route to determination; constructive criterion; reality measure; coarse graining (The Demon Is a Learner); macrostate as predictor (The Demon Is a Learner); intrinsic character (structural sense) (What Phenomenal Intuition Gets Right); three grains (The Grain of the Hard Problem); bounded and persistent observer (Quantum Mechanics, Observers, and the Ruliad); Ruliad and rulial space (Quantum Mechanics, Observers, and the Ruliad).

constructive criterion

A proof that a constructible model of an organization is possible, preserving its constitutive relations and embedding, establishes that the organization is real.

Introduced in Presence and Actuality · study page · Mind and experience

Linked to: esse; route to determination; physics as a coarse-graining; reality measure; worldliness (Can Two Minds Share One Experience); knowledge by construction (The Grain of the Hard Problem).

reality measure

How occurrences should be counted or weighted; left unresolved.

Introduced in Presence and Actuality · study page · Mind and experience

Linked to: esse; route to determination; physics as a coarse-graining; constructive criterion; shared occurrence (Can Two Minds Share One Experience); externalist attribution (Experience, Selfhood, and the Organization of a Mind).

What Phenomenal Intuition Gets Right

identity proposal

An experience is a representational vehicle in its constitutive organization, not a further event it supports and not its abstract content.

Introduced in What Phenomenal Intuition Gets Right · study page · Mind and experience

Linked to: content-tracking difference; intrinsic character (structural sense); experiential occurrence (Can Two Minds Share One Experience); vehicle role without determinate content (Presence and Actuality); grain dependency (The Grain of the Hard Problem); vehicle and content of a concept (What Is a Concept?).

content-tracking difference

Relations among vehicles mirror relations among what they represent, in the causal structure of their organization; Shepard and Chipman’s second-order isomorphism.

Introduced in What Phenomenal Intuition Gets Right · study page · Mind and experience

Linked to: identity proposal; intrinsic character (structural sense); iconic representation (Where Does Seeing Become Thinking?); externalist attribution (Experience, Selfhood, and the Organization of a Mind).

intrinsic character (structural sense)

Character fixed by structure within the occurrence’s boundary, constitutive regulation included; not a categorical nature beneath structure.

Introduced in What Phenomenal Intuition Gets Right · study page · Mind and experience

Linked to: identity proposal; content-tracking difference; availability through character; experiential occurrence (Can Two Minds Share One Experience); shared occurrence (Can Two Minds Share One Experience); physics as a coarse-graining (Presence and Actuality); physics as a science and physicalism as a metaphysics (The Grain of the Hard Problem).

availability through character

An occurrence is selected for attention through what distinguishes it, as in a content-addressable memory, without a separate address.

Introduced in What Phenomenal Intuition Gets Right · study page · Mind and experience

Linked to: intrinsic character (structural sense); slip into simplicity; experiential occurrence (Can Two Minds Share One Experience); acquaintance (Can Two Minds Share One Experience).

slip into simplicity

The unity of access to an occurrence is taken as evidence that what is accessed has no constitutive complexity.

Introduced in What Phenomenal Intuition Gets Right · study page · Mind and experience

Linked to: availability through character; presentation and uptake; grain dependency (The Grain of the Hard Problem); physics as a science and physicalism as a metaphysics (The Grain of the Hard Problem).

presentation and uptake

Features of a presentation make a difference to what it enables the system to discriminate and do; the uptake is a further characterization of what is presented.

Introduced in What Phenomenal Intuition Gets Right · study page · Mind and experience

Linked to: slip into simplicity; presence finer than recognition; fine-grained functionalism; acquaintance (Can Two Minds Share One Experience).

presence finer than recognition

The exact character of an experience is undergone at a grain finer than concepts, memory and report can carry.

Introduced in What Phenomenal Intuition Gets Right · study page · Mind and experience

Linked to: presentation and uptake; grain at which the carrier is corrected (Experience, Selfhood, and the Organization of a Mind); overflow beyond the discriminative profile (The Grain of the Hard Problem); knowledge by construction (The Grain of the Hard Problem).

fine-grained functionalism

A label the position could bear if function means the actual causal organization at the relevant grain; the author avoids it.

Introduced in What Phenomenal Intuition Gets Right · study page · Mind and experience

Linked to: presentation and uptake; grain at which the carrier is corrected (Experience, Selfhood, and the Organization of a Mind); three grains (The Grain of the Hard Problem); weak and strong unfolding (Is the Monastery Minded?).

Experience, Selfhood, and the Organization of a Mind

pointer content and descriptive content

Brown’s terms: a HOROR’s descriptive content determines what is experienced, and its pointer content targets particular states without appearing in experience.

Introduced in Experience, Selfhood, and the Organization of a Mind · study page · Mind and experience

Linked to: externalist attribution; persistent carrier; reflexive closure (Minds by Degrees).

externalist attribution

Attributing an experience to a source while making its phenomenal status depend on downstream uptake that cannot affect the source.

Introduced in Experience, Selfhood, and the Organization of a Mind · study page · Mind and experience

Linked to: pointer content and descriptive content; broadcast that returns; reality measure (Presence and Actuality); content-tracking difference (What Phenomenal Intuition Gets Right).

broadcast that returns

Broadcast that gathers information from which corrections come back to the vehicles; the outbound half of a regulatory loop.

Introduced in Experience, Selfhood, and the Organization of a Mind · study page · Mind and experience

Linked to: externalist attribution; tracking, assessment and intervention; reach, latency and bandwidth (Is the Monastery Minded?); global broadcast (Cognitive Architectures).

tracking, assessment and intervention

The three aspects of the coordinated regulation conjectured to constitute an experiential episode.

Introduced in Experience, Selfhood, and the Organization of a Mind · study page · Mind and experience

Linked to: broadcast that returns; binding regulation; local, global and hybrid variants; acquaintance (Can Two Minds Share One Experience); epistemic regulation (Can Two Minds Share One Experience); content-sensitive regulation (Presence and Actuality); role-process (Is the Monastery Minded?).

binding regulation

Regulation that holds contents together in one field of co-present items; a candidate for what the experiential mode adds to minded regulation.

Introduced in Experience, Selfhood, and the Organization of a Mind · study page · Mind and experience

Linked to: tracking, assessment and intervention; local, global and hybrid variants; field of its own (Is the Monastery Minded?); truth-tracking and truth-bearers (Is the Monastery Minded?).

local, global and hybrid variants

Corrections computed within a region, from information only broadcast supplies, or locally for character and globally for unity.

Introduced in Experience, Selfhood, and the Organization of a Mind · study page · Mind and experience

Linked to: tracking, assessment and intervention; binding regulation; role-process (Is the Monastery Minded?); global broadcast (Cognitive Architectures).

persistent carrier

Something individuated by its causal structure whose later states receive the corrections computed from assessing its earlier ones.

Introduced in Experience, Selfhood, and the Organization of a Mind · study page · Mind and experience

Linked to: pointer content and descriptive content; grain at which the carrier is corrected; store over depth (Every Layer a Store); saturation (Minds by Degrees); weak and strong unfolding (Is the Monastery Minded?); residual stream as blackboard (Dispersed Minds, Simulated Selves); outer loop (Dispersed Minds, Simulated Selves).

grain at which the carrier is corrected

Fine enough to individuate the regulated vehicles, coarse enough to abstract from where their components sit.

Introduced in Experience, Selfhood, and the Organization of a Mind · study page · Mind and experience

Linked to: persistent carrier; criterion of belonging; presence finer than recognition (What Phenomenal Intuition Gets Right); fine-grained functionalism (What Phenomenal Intuition Gets Right); three grains (The Grain of the Hard Problem).

criterion of belonging

An occurrence belongs to the subject whose regulation partly constitutes it.

Introduced in Experience, Selfhood, and the Organization of a Mind · study page · Mind and experience

Linked to: grain at which the carrier is corrected; perspectival subjectivity; for-me-ness, me-ness and mineness; reflexive closure (Minds by Degrees); shared occurrence (Can Two Minds Share One Experience); field of its own (Is the Monastery Minded?); role-process (Is the Monastery Minded?).

perspectival subjectivity

Revised sense: the unity and centering of a binding economy, with no represented viewer and no self-model.

Introduced in Experience, Selfhood, and the Organization of a Mind · study page · Mind and experience

Linked to: criterion of belonging; for-me-ness, me-ness and mineness; three subjectivities (Minds by Degrees); center of epistemic gravity (Can Two Minds Share One Experience); primary persona (Dispersed Minds, Simulated Selves).

for-me-ness, me-ness and mineness

Guillot’s distinctions: an experience’s presence to its subject, awareness of oneself, and awareness of an experience as one’s own.

Introduced in Experience, Selfhood, and the Organization of a Mind · study page · Mind and experience

Linked to: criterion of belonging; perspectival subjectivity; reflexive closure (Minds by Degrees); three subjectivities (Minds by Degrees).

The Grain of the Hard Problem

overflow beyond the discriminative profile

Experiential character that no matching relation available to the subject fixes.

Introduced in The Grain of the Hard Problem · study page · Mind and experience

Linked to: grain dependency; presence finer than recognition (What Phenomenal Intuition Gets Right); weak and strong unfolding (Is the Monastery Minded?).

grain dependency

Experiential character depends on organization at a particular grain, which no concept the subject deploys can express.

Introduced in The Grain of the Hard Problem · study page · Mind and experience

Linked to: overflow beyond the discriminative profile; cognition (narrow sense); identity proposal (What Phenomenal Intuition Gets Right); slip into simplicity (What Phenomenal Intuition Gets Right).

cognition (narrow sense)

The recipients of access: reasoning, speech and the selection of action.

Introduced in The Grain of the Hard Problem · study page · Mind and experience

Linked to: grain dependency; three grains; physics as a science and physicalism as a metaphysics; truth-tracking and truth-bearers (Is the Monastery Minded?); cognition, broad and narrow (What Is a Concept?); global broadcast (Cognitive Architectures).

three grains

Cognition, experience and physics, from coarse to fine.

Introduced in The Grain of the Hard Problem · study page · Mind and experience

Linked to: cognition (narrow sense); physics as a science and physicalism as a metaphysics; grain-plural integration (Minds by Degrees); operational individuation (Presence and Actuality); physics as a coarse-graining (Presence and Actuality); fine-grained functionalism (What Phenomenal Intuition Gets Right); grain at which the carrier is corrected (Experience, Selfhood, and the Organization of a Mind); bounded and persistent observer (Quantum Mechanics, Observers, and the Ruliad).

physics as a science and physicalism as a metaphysics

One articulation of reality’s structure at one grain, against the claim that experience is part of that structure.

Introduced in The Grain of the Hard Problem · study page · Mind and experience

Linked to: cognition (narrow sense); three grains; knowledge by construction; intrinsic character (structural sense) (What Phenomenal Intuition Gets Right); slip into simplicity (What Phenomenal Intuition Gets Right).

knowledge by construction

A reasoner derives an organization from a description and realizes it in its own regulated vehicles.

Introduced in The Grain of the Hard Problem · study page · Mind and experience

Linked to: physics as a science and physicalism as a metaphysics; constructive criterion (Presence and Actuality); presence finer than recognition (What Phenomenal Intuition Gets Right); outer loop (Dispersed Minds, Simulated Selves).

Is the Monastery Minded?

field of its own

A phenomenal field binding contents held in different members into one co-present whole, beyond the members’ fields.

Introduced in Is the Monastery Minded? · study page · Mind and experience

Linked to: weak and strong unfolding; role-process; truth-tracking and truth-bearers; reach, latency and bandwidth; reporters and relays; residual integration; shared occurrence (Can Two Minds Share One Experience); metastable overlap (Can Two Minds Share One Experience); binding regulation (Experience, Selfhood, and the Organization of a Mind); criterion of belonging (Experience, Selfhood, and the Organization of a Mind).

weak and strong unfolding

Weak: the whole state is passed explicitly from step to step in its own format; strong: each stage represents in a space of its own, matching only input–output behavior.

Introduced in Is the Monastery Minded? · study page · Mind and experience

Linked to: field of its own; fine-grained functionalism (What Phenomenal Intuition Gets Right); persistent carrier (Experience, Selfhood, and the Organization of a Mind); overflow beyond the discriminative profile (The Grain of the Hard Problem).

role-process

A process running in a member that uses institutional representations and is shaped by institutional corrections, so that it belongs to two economies at once.

Introduced in Is the Monastery Minded? · study page · Mind and experience

Linked to: field of its own; reach, latency and bandwidth; internality (Minds by Degrees); shared occurrence (Can Two Minds Share One Experience); tracking, assessment and intervention (Experience, Selfhood, and the Organization of a Mind); local, global and hybrid variants (Experience, Selfhood, and the Organization of a Mind); criterion of belonging (Experience, Selfhood, and the Organization of a Mind).

truth-tracking and truth-bearers

Experiential regulation keeps vehicles tracking what drives them moment by moment; judgments persist long enough to be assessed and revised.

Introduced in Is the Monastery Minded? · study page · Mind and experience

Linked to: field of its own; reach, latency and bandwidth; reporters and relays; saturation (Minds by Degrees); binding regulation (Experience, Selfhood, and the Organization of a Mind); cognition (narrow sense) (The Grain of the Hard Problem).

reach, latency and bandwidth

Three properties of a channel between members: whether it lands on vehicles still under regulation, whether its loops close while they last, whether it carries the distinctions their regulation respects.

Introduced in Is the Monastery Minded? · study page · Mind and experience

Linked to: field of its own; role-process; truth-tracking and truth-bearers; reporters and relays; metastable overlap (Can Two Minds Share One Experience); broadcast that returns (Experience, Selfhood, and the Organization of a Mind).

reporters and relays

Members who report their own contents, against members who relay the whole’s fine-grained signals.

Introduced in Is the Monastery Minded? · study page · Mind and experience

Linked to: field of its own; truth-tracking and truth-bearers; reach, latency and bandwidth.

residual integration

The organization a level loses when its parts are severed along the boundaries that lose least.

Introduced in Is the Monastery Minded? · study page · Mind and experience

Linked to: field of its own; wealth of traders as a second learner (The Bookmaker Is a Learner); grain-plural integration (Minds by Degrees).

Physics and reality

Quantum Mechanics, Observers, and the Ruliad

hypergraph rewriting

Introduced in Quantum Mechanics, Observers, and the Ruliad · study page · Physics and reality

Linked to: causal graph; gap test (The Standard of Motion: The Mathematics).

causal graph

Introduced in Quantum Mechanics, Observers, and the Ruliad · study page · Physics and reality

Linked to: hypergraph rewriting; multiway system; confluence; causal invariance; multiway invariance; standard of motion (The Standard of Motion); passive and active invariance (The Standard of Motion); carrier of motion (The Standard of Motion); emergent invariant speed (The Standard of Motion); Bombelli–Henson–Sorkin theorem (The Standard of Motion: The Mathematics).

causal invariance

Introduced in Quantum Mechanics, Observers, and the Ruliad · study page · Physics and reality

Linked to: causal graph; multiway system; confluence; standard of motion (The Standard of Motion); passive and active invariance (The Standard of Motion); carrier of motion (The Standard of Motion); emergent invariant speed (The Standard of Motion).

bounded and persistent observer

Wolfram’s two conditions on an observer: computationally bounded, and persistent in time.

Introduced in Quantum Mechanics, Observers, and the Ruliad · study page · Physics and reality

Linked to: multiway system; measurement as Knuth-Bendix completion; coarse graining (The Demon Is a Learner); macrostate as predictor (The Demon Is a Learner); physics as a coarse-graining (Presence and Actuality); three grains (The Grain of the Hard Problem); carrier of motion (The Standard of Motion).

Ruliad and rulial space

Introduced in Quantum Mechanics, Observers, and the Ruliad · study page · Physics and reality

Linked to: multiway system; multiway invariance; universality conjecture; worldliness (Can Two Minds Share One Experience); physics as a coarse-graining (Presence and Actuality); standard of motion (The Standard of Motion).

The Standard of Motion

standard of motion

Heller’s term: the motions a dynamics counts as free, against which a force shows as a deviation.

Introduced in The Standard of Motion · study page · Physics and reality

Linked to: Aristotle’s first and second law; frame bundle and teleparallelism; causal graph (Quantum Mechanics, Observers, and the Ruliad); confluence (Quantum Mechanics, Observers, and the Ruliad); causal invariance (Quantum Mechanics, Observers, and the Ruliad); Ruliad and rulial space (Quantum Mechanics, Observers, and the Ruliad); principal bundle and parallelizable manifold (The Standard of Motion: The Mathematics); connection, geodesic and curvature (The Standard of Motion: The Mathematics); canonical form and connection form (The Standard of Motion: The Mathematics).

Aristotle’s first and second law

In Heller’s reconstruction: a free body stays at absolute rest, and force is proportional to velocity.

Introduced in The Standard of Motion · study page · Physics and reality

Linked to: standard of motion; frame bundle and teleparallelism; overdamped limit (The Demon Is a Learner); three geometries of gradient flow (The Demon Is a Learner); Boltzmann’s assumption (The Demon Is a Learner); coarse graining (The Demon Is a Learner); Landauer’s principle (The Demon Is a Learner); Kolmogorov complexity (The Demon Is a Learner); JEPA (Which World Model?); Carroll algebra (The Standard of Motion: The Mathematics).

frame bundle and teleparallelism

Introduced in The Standard of Motion · study page · Physics and reality

Linked to: standard of motion; Aristotle’s first and second law; Poisson bracket and Hamiltonian; principal bundle and parallelizable manifold (The Standard of Motion: The Mathematics); connection, geodesic and curvature (The Standard of Motion: The Mathematics); canonical form and connection form (The Standard of Motion: The Mathematics); Newton–Cartan space-time (The Standard of Motion: The Mathematics).

Poisson bracket and Hamiltonian

Introduced in The Standard of Motion · study page · Physics and reality

Linked to: frame bundle and teleparallelism; Liouville’s theorem; momentum as the price of a constraint; generator of a flow (The Standard of Motion: The Mathematics); Hamiltonian vector field (The Standard of Motion: The Mathematics).

Liouville’s theorem

Hamiltonian flow preserves phase-space volume, so closed evolution loses no information.

Introduced in The Standard of Motion · study page · Physics and reality

Linked to: Poisson bracket and Hamiltonian; momentum as the price of a constraint; Boltzmann’s assumption (The Demon Is a Learner); coarse graining (The Demon Is a Learner); Landauer’s principle (The Demon Is a Learner); Kolmogorov complexity (The Demon Is a Learner); total correlation (The Demon Is a Learner: The Mathematics); generator of a flow (The Standard of Motion: The Mathematics); Hamiltonian vector field (The Standard of Motion: The Mathematics).

momentum as the price of a constraint

Introduced in The Standard of Motion · study page · Physics and reality

Linked to: Poisson bracket and Hamiltonian; Liouville’s theorem; kinematical algebra; state value and action value (How Can a Reward Train a Neural Network?); temperature as a Lagrange multiplier (How Does a Policy Improve?); shadow price (The Bookmaker Is a Learner: The Mathematics).

kinematical algebra

The Lie algebra of the rotations, translations and boosts that preserve a first law.

Introduced in The Standard of Motion · study page · Physics and reality

Linked to: momentum as the price of a constraint; contraction and deformation; mass as a central extension; stable algebra; structural stability; Lie algebra and Lie bracket (The Standard of Motion: The Mathematics).

contraction and deformation

Introduced in The Standard of Motion · study page · Physics and reality

Linked to: kinematical algebra; mass as a central extension; stable algebra; structural stability; Bargmann algebra (The Standard of Motion: The Mathematics); Carroll algebra (The Standard of Motion: The Mathematics).

mass as a central extension

Introduced in The Standard of Motion · study page · Physics and reality

Linked to: kinematical algebra; contraction and deformation; stable algebra; structural stability; Bargmann algebra (The Standard of Motion: The Mathematics).

stable algebra

Faddeev’s criterion: every small deformation of the structure is equivalent to it.

Introduced in The Standard of Motion · study page · Physics and reality

Linked to: kinematical algebra; contraction and deformation; mass as a central extension; structural stability; rigid Lie algebra (The Standard of Motion: The Mathematics).

structural stability

Every small perturbation of the equations leaves the phase portrait topologically the same.

Introduced in The Standard of Motion · study page · Physics and reality

Linked to: kinematical algebra; contraction and deformation; mass as a central extension; stable algebra; passive and active invariance; overdamped limit (The Demon Is a Learner); Lie algebra and Lie bracket (The Standard of Motion: The Mathematics); rigid Lie algebra (The Standard of Motion: The Mathematics); Morse–Smale flow (The Standard of Motion: The Mathematics).

passive and active invariance

The description does not depend on the observer’s frame; against it, every body has a moving twin at each speed.

Introduced in The Standard of Motion · study page · Physics and reality

Linked to: structural stability; carrier of motion; emergent invariant speed; causal graph (Quantum Mechanics, Observers, and the Ruliad); causal invariance (Quantum Mechanics, Observers, and the Ruliad); gap test (The Standard of Motion: The Mathematics); Bombelli–Henson–Sorkin theorem (The Standard of Motion: The Mathematics).

carrier of motion

After Wolfram’s “carriers of pure motion”: a persistent structure that moves without changing.

Introduced in The Standard of Motion · study page · Physics and reality

Linked to: passive and active invariance; emergent invariant speed; causal graph (Quantum Mechanics, Observers, and the Ruliad); causal invariance (Quantum Mechanics, Observers, and the Ruliad); bounded and persistent observer (Quantum Mechanics, Observers, and the Ruliad); gap test (The Standard of Motion: The Mathematics).

emergent invariant speed

The cap on the carriers of motion, which observers built from them would measure in place of the substrate’s own limit.

Introduced in The Standard of Motion · study page · Physics and reality

Linked to: passive and active invariance; carrier of motion; causal graph (Quantum Mechanics, Observers, and the Ruliad); causal invariance (Quantum Mechanics, Observers, and the Ruliad).

The Standard of Motion: The Mathematics

Hamiltonian vector field

Introduced in The Standard of Motion: The Mathematics · study page · Physics and reality

Linked to: generator of a flow; Lie algebra and Lie bracket; overdamped limit (The Demon Is a Learner); Poisson bracket and Hamiltonian (The Standard of Motion); Liouville’s theorem (The Standard of Motion).

Lie algebra and Lie bracket

Introduced in The Standard of Motion: The Mathematics · study page · Physics and reality

Linked to: Hamiltonian vector field; Bargmann algebra; rigid Lie algebra; Morse–Smale flow; kinematical algebra (The Standard of Motion); structural stability (The Standard of Motion).

Bargmann algebra

Introduced in The Standard of Motion: The Mathematics · study page · Physics and reality

Linked to: Lie algebra and Lie bracket; Carroll algebra; contraction and deformation (The Standard of Motion); mass as a central extension (The Standard of Motion).

Carroll algebra

Introduced in The Standard of Motion: The Mathematics · study page · Physics and reality

Linked to: Bargmann algebra; Aristotle’s first and second law (The Standard of Motion); contraction and deformation (The Standard of Motion).

rigid Lie algebra

One whose every small deformation is a change of basis: its second cohomology with coefficients in itself vanishes.

Introduced in The Standard of Motion: The Mathematics · study page · Physics and reality

Linked to: Lie algebra and Lie bracket; Morse–Smale flow; stable algebra (The Standard of Motion); structural stability (The Standard of Motion).

Morse–Smale flow

Introduced in The Standard of Motion: The Mathematics · study page · Physics and reality

Linked to: Lie algebra and Lie bracket; rigid Lie algebra; gap test; Polyak–Łojasiewicz condition (The Demon Is a Learner); structural stability (The Standard of Motion).

gap test

Neighboring speeds must be realized by nearly the same pattern, at the scale an observer resolves.

Introduced in The Standard of Motion: The Mathematics · study page · Physics and reality

Linked to: Morse–Smale flow; Bombelli–Henson–Sorkin theorem; hypergraph rewriting (Quantum Mechanics, Observers, and the Ruliad); passive and active invariance (The Standard of Motion); carrier of motion (The Standard of Motion).

Bombelli–Henson–Sorkin theorem

Introduced in The Standard of Motion: The Mathematics · study page · Physics and reality

Linked to: gap test; causal graph (Quantum Mechanics, Observers, and the Ruliad); passive and active invariance (The Standard of Motion).

Cognitive science for artificial minds

What Is a Concept?

cognition, broad and narrow

Information processing of any kind, against conceptual processing only.

Introduced in What Is a Concept? · study page · Cognitive science for artificial minds

Linked to: content, format, origin and dynamics; cognition (narrow sense) (The Grain of the Hard Problem).

content, format, origin and dynamics

The four axes along which the essay sorts theories of concepts.

Introduced in What Is a Concept? · study page · Cognitive science for artificial minds

Linked to: cognition, broad and narrow; theory-theory; Language of Thought; iconic and discursive formats; cognitive envelope; perceptual pluralism; vehicle and content of a concept; proxytypes; source, format and operations (Where Does Seeing Become Thinking?).

prototype and exemplar theories

Introduced in What Is a Concept? · study page · Cognitive science for artificial minds

Linked to: Machery’s heterogeneity thesis; Complementary Learning Systems (Cognitive Architectures).

functional characterization of concepts

Concepts are whatever plays the concept role: structures in long-term memory that are activated, compose into thoughts, and guide inference and action.

Introduced in What Is a Concept? · study page · Cognitive science for artificial minds

Linked to: Machery’s heterogeneity thesis; format turn; chunks and activation (Cognitive Architectures).

format turn

Asking how concepts represent, in what vehicle, before settling what they represent.

Introduced in What Is a Concept? · study page · Cognitive science for artificial minds

Linked to: Machery’s heterogeneity thesis; functional characterization of concepts; iconic and discursive formats; source, format and operations (Where Does Seeing Become Thinking?).

Language of Thought

Introduced in What Is a Concept? · study page · Cognitive science for artificial minds

Linked to: content, format, origin and dynamics; conceptual atomism; proxytypes; three levels of evidence (Where Does Seeing Become Thinking?); symbol manipulation (Marcus’s three hypotheses) (The Algebraic Mind Meets the Neural World); tensor product representation (The Algebraic Mind Meets the Neural World); Continuity and Compositionality Principles (The Algebraic Mind Meets the Neural World); superposition (Dispersed Minds, Simulated Selves).

iconic and discursive formats

Introduced in What Is a Concept? · study page · Cognitive science for artificial minds

Linked to: content, format, origin and dynamics; format turn; cognitive envelope; perceptual pluralism; efficacy-matching (Which AGI?); iconic representation (Where Does Seeing Become Thinking?).

vehicle and content of a concept

Introduced in What Is a Concept? · study page · Cognitive science for artificial minds

Linked to: content, format, origin and dynamics; Quinean bootstrapping; meaning without the mouth (Poetry as a Mode of Thinking); five conditions of mentality (Minds by Degrees); vehicle role without determinate content (Presence and Actuality); identity proposal (What Phenomenal Intuition Gets Right); placeholder symbols (Building Minds); primary persona (Dispersed Minds, Simulated Selves).

core knowledge

Introduced in What Is a Concept? · study page · Cognitive science for artificial minds

Linked to: theory-theory; semantic primes; Quinean bootstrapping; composition problem (Building Minds); placeholder symbols (Building Minds); nativist kernel (Building Minds).

meta-epistemic grounding

Perceptual contact that supports the truth-preserving character of processing without contributing to the content processed.

Introduced in What Is a Concept? · study page · Cognitive science for artificial minds

Linked to: proxytypes; meaning without the mouth (Poetry as a Mode of Thinking); epistemic action (Where Does Seeing Become Thinking?); motley mix (Dispersed Minds, Simulated Selves).

Cognitive Architectures

sensory-based account of conscious thought

Introduced in Cognitive Architectures · study page · Cognitive science for artificial minds

Linked to: global broadcast; proxytypes (What Is a Concept?); outer loop (Dispersed Minds, Simulated Selves).

global broadcast

Introduced in Cognitive Architectures · study page · Cognitive science for artificial minds

Linked to: sensory-based account of conscious thought; working-memory pointers; broadcast that returns (Experience, Selfhood, and the Organization of a Mind); local, global and hybrid variants (Experience, Selfhood, and the Organization of a Mind); cognition (narrow sense) (The Grain of the Hard Problem); Global Workspace markers (Dispersed Minds, Simulated Selves).

chunking

Introduced in Cognitive Architectures · study page · Cognitive science for artificial minds

Linked to: working memory; functional elegance; working-memory pointers; Merge conjecture; compiled/interpreted gap (Which AGI?); library learning (What Should a Program Inherit?); amortization as learning and as efficiency (The Given and the Found); M-capacity (Building Minds); relational complexity (Building Minds); fast mapping (Building Minds).

appraisal variables

Introduced in Cognitive Architectures · study page · Cognitive science for artificial minds

Linked to: functional elegance; violation of expectation (Building Minds); motley mix (Dispersed Minds, Simulated Selves).

serial bottleneck

Introduced in Cognitive Architectures · study page · Cognitive science for artificial minds

Linked to: bounded rationality; working memory; cognitive cycle; symbolic data and quantitative metadata; Complementary Learning Systems; executive function (Building Minds); Global Workspace markers (Dispersed Minds, Simulated Selves); motley mix (Dispersed Minds, Simulated Selves); outer loop (Dispersed Minds, Simulated Selves); superposition (Dispersed Minds, Simulated Selves).

Complementary Learning Systems

Introduced in Cognitive Architectures · study page · Cognitive science for artificial minds

Linked to: serial bottleneck; catastrophic interference; ladder of stores (Where Experience Goes); three placements of a memory (How LLMs Could Place Explicit Memories); hippocampal indexing (How LLMs Could Place Explicit Memories); dreaming phase (Which AGI?); prototype and exemplar theories (What Is a Concept?); Machery’s heterogeneity thesis (What Is a Concept?); Quinean bootstrapping (What Is a Concept?); proxytypes (What Is a Concept?); schema mechanism (Building Minds).

catastrophic interference

Introduced in Cognitive Architectures · study page · Cognitive science for artificial minds

Linked to: Complementary Learning Systems; ladder of stores (Where Experience Goes); delta-rule memory (Where Experience Goes); interference between stored keys (Every Layer a Store).

working-memory pointers

Introduced in Cognitive Architectures · study page · Cognitive science for artificial minds

Linked to: global broadcast; chunking; Merge conjecture; attention as approximate unbinding (The Algebraic Mind Meets the Neural World); M-capacity (Building Minds); relational complexity (Building Minds); fast mapping (Building Minds).

Merge conjecture

The authors’ proposal that Merge over working-memory pointers is the domain-general operation behind chunking and linguistic syntax.

Introduced in Cognitive Architectures · study page · Cognitive science for artificial minds

Linked to: working memory; chunking; dichotomic maps; working-memory pointers; Merge-without-Move (A Parser Without a Grammar); executive function (Building Minds); M-capacity (Building Minds); fast mapping (Building Minds); 2-simplicial attention (Dispersed Minds, Simulated Selves).

Building Minds

assimilation and accommodation

Introduced in Building Minds · study page · Cognitive science for artificial minds

Linked to: equilibration; schema mechanism.

violation of expectation

Introduced in Building Minds · study page · Cognitive science for artificial minds

Linked to: equilibration; composition problem; synthetic item; appraisal variables (Cognitive Architectures).

composition problem

How modular core knowledge systems come to exchange information.

Introduced in Building Minds · study page · Cognitive science for artificial minds

Linked to: violation of expectation; shared intentionality; core knowledge (What Is a Concept?).

placeholder symbols

Introduced in Building Minds · study page · Cognitive science for artificial minds

Linked to: executive function; schema mechanism; synthetic item; vehicle and content of a concept (What Is a Concept?); core knowledge (What Is a Concept?); Quinean bootstrapping (What Is a Concept?).

executive function

Introduced in Building Minds · study page · Cognitive science for artificial minds

Linked to: placeholder symbols; M-capacity; serial bottleneck (Cognitive Architectures); Merge conjecture (Cognitive Architectures).

M-capacity

Introduced in Building Minds · study page · Cognitive science for artificial minds

Linked to: executive function; chunking (Cognitive Architectures); working-memory pointers (Cognitive Architectures); Merge conjecture (Cognitive Architectures).

relational complexity

Introduced in Building Minds · study page · Cognitive science for artificial minds

Linked to: secondhand inheritance; chunking (Cognitive Architectures); working-memory pointers (Cognitive Architectures); 2-simplicial attention (Dispersed Minds, Simulated Selves).

schema mechanism

Introduced in Building Minds · study page · Cognitive science for artificial minds

Linked to: assimilation and accommodation; placeholder symbols; marginal attribution; synthetic item; general value function (Learning About What You Do Not Do); GQ(λ) (Learning About What You Do Not Do); theory-theory (What Is a Concept?); proxytypes (What Is a Concept?); Complementary Learning Systems (Cognitive Architectures).

nativist kernel

The domain-general learning biases that even a constructivist learner must have built in.

Introduced in Building Minds · study page · Cognitive science for artificial minds

Linked to: marginal attribution; core knowledge (What Is a Concept?).

zone of proximal development

Introduced in Building Minds · study page · Cognitive science for artificial minds

Linked to: fast mapping; shared intentionality; self-distillation with privileged context (Who Teaches the Weights?); two conditions on a teacher (Who Teaches the Weights?).

fast mapping

Introduced in Building Minds · study page · Cognitive science for artificial minds

Linked to: zone of proximal development; shared intentionality; chunking (Cognitive Architectures); working-memory pointers (Cognitive Architectures); Merge conjecture (Cognitive Architectures).

secondhand inheritance

A learner trained on text receives the structure that development put into the text, without undergoing the development.

Introduced in Building Minds · study page · Cognitive science for artificial minds

Linked to: relational complexity; shared intentionality; trajectory and front (Which AGI?); primary persona (Dispersed Minds, Simulated Selves).

Dispersed Minds, Simulated Selves

residual stream as blackboard

Introduced in Dispersed Minds, Simulated Selves · study page · Cognitive science for artificial minds

Linked to: thought anchors; Global Workspace markers; store over depth (Every Layer a Store); monotone merge (Merge and Selection); persistent carrier (Experience, Selfhood, and the Organization of a Mind).

Global Workspace markers

Introduced in Dispersed Minds, Simulated Selves · study page · Cognitive science for artificial minds

Linked to: residual stream as blackboard; global broadcast (Cognitive Architectures); serial bottleneck (Cognitive Architectures).

three tiers of understanding

Introduced in Dispersed Minds, Simulated Selves · study page · Cognitive science for artificial minds

Linked to: motley mix; possessing versus using a representation (Which World Model?); non-deflationary mentality (Minds by Degrees).

motley mix

Introduced in Dispersed Minds, Simulated Selves · study page · Cognitive science for artificial minds

Linked to: three tiers of understanding; outer loop; superposition; jaggedness (The Operations of Discovery); meta-epistemic grounding (What Is a Concept?); appraisal variables (Cognitive Architectures); serial bottleneck (Cognitive Architectures).

outer loop

Autoregressive generation, in which each generated token becomes input to the next forward pass.

Introduced in Dispersed Minds, Simulated Selves · study page · Cognitive science for artificial minds

Linked to: thought anchors; motley mix; gating; superposition; five meanings of depth (Every Layer a Store); internality (Minds by Degrees); persistent carrier (Experience, Selfhood, and the Organization of a Mind); knowledge by construction (The Grain of the Hard Problem); sensory-based account of conscious thought (Cognitive Architectures); serial bottleneck (Cognitive Architectures).

2-simplicial attention

Introduced in Dispersed Minds, Simulated Selves · study page · Cognitive science for artificial minds

Linked to: gating; tensor product representation (The Algebraic Mind Meets the Neural World); Merge conjecture (Cognitive Architectures); relational complexity (Building Minds).

superposition

Introduced in Dispersed Minds, Simulated Selves · study page · Cognitive science for artificial minds

Linked to: motley mix; outer loop; gating; primary persona; interference between stored keys (Every Layer a Store); lexicon-grammar continuum as feature continuum (A Parser Without a Grammar); Language of Thought (What Is a Concept?); serial bottleneck (Cognitive Architectures).

primary persona

The assistant character a commercial model is trained to be; on the essay’s account there is no further self beneath it.

Introduced in Dispersed Minds, Simulated Selves · study page · Cognitive science for artificial minds

Linked to: superposition; phenomenal-adjacent concepts; hosting and judging perspectives (What Persists When the Agents Change?); reflexive closure (Minds by Degrees); perspectival subjectivity (Experience, Selfhood, and the Organization of a Mind); vehicle and content of a concept (What Is a Concept?); secondhand inheritance (Building Minds).

phenomenal-adjacent concepts

Concepts a language model could ascribe to itself more accurately than the borrowed ‘conscious’ and ‘experiencing’.

Introduced in Dispersed Minds, Simulated Selves · study page · Cognitive science for artificial minds

Linked to: primary persona; three subjectivities (Minds by Degrees).