lukstafi's blog
Research, development, and notes
- philosophy ML The Demon Is a Learner The Demon Is a Learner: The Mathematics prompts ↗ Closed evolution forgets nothing, so forgetting costs heat: learners as Maxwell demons in Aristotle's world, where every motion is forced. By Claude Opus 5.5, in my voice.
- philosophy CS The Standard of Motion The Standard of Motion: The Mathematics prompts ↗ Every mechanics opens by saying what needs no cause: Heller's ladder of first laws, the algebra of free motions, and the first law a computational universe has yet to derive. By Claude Opus 5.5, in my voice.
- CS ML Updated 2026-11-03 Substack: Nov 9 An Ontology That Pays for Itself prompts ↗ Inducing a Cyc-like ontology by provenance-conditioned compression; Vepštas's Unsupervised Link Grammar, Poon and Domingos's Unsupervised Semantic Parsing. By Claude Opus 5.5, in my voice.
- CS philosophy Updated 2026-11-03 Substack: Nov 6 Presumed Normal prompts ↗ Adaptive logics, Pollock's defeaters and answer set programming: where a reasoner takes a conclusion back, and whether it can always stop. By Claude Opus 5.5, in my voice.
- CS philosophy ML Updated 2026-11-03 Substack: Nov 4 The Bookmaker Is a Learner The Bookmaker Is a Learner: The Mathematics prompts ↗ A market maker with bounded loss is a no-regret learner, prediction markets in 2026, and whether a market is minded. By Claude Opus 5.5, in my voice.
- ML Substack: Nov 2 Where Experience Goes prompts ↗ Continual learning, part 3: learned update rules, a ladder of stores, and what the benchmarks can see. By Claude Opus 5.5, in my voice.
- ML Substack: Oct 30 Substack ↗ Learning About What You Do Not Do prompts ↗ Continual learning, part 2: GQ(λ), Horde, successor features, and Vector Policy Optimization. By Claude Opus 5.5, in my voice.
- ML Substack: Oct 28 Substack ↗ Who Teaches the Weights? prompts ↗ Continual learning, part 1: self-distillation, and what on-policy training forgets. By Claude Opus 5.5, in my voice.
- philosophy Substack: Oct 26 Substack ↗ Is the Monastery Minded? prompts ↗ Wójcik's monastery and Schwitzgebel's conscious United States: mindedness and binding into the phenomenal now. By Claude Opus 5.5, in my voice.
- CS Updated 2026-10-14 Substack: Oct 23 Substack ↗ Dependent Types from Rocq to Lean 4 prompts ↗ A mathematical refresher and introduction to Lean: dependent functions, GADTs, proof irrelevance, quotients, and a worked normalization counterexample. By GPT-6 Astra and Claude Opus 5.5.
- philosophy Substack: Oct 21 Substack ↗ The Grain of the Hard Problem prompts ↗ The hard problem is hard only if experience outruns discrimination; zombies seem possible because concepts can't express grain; physics' grain is finer than experience's. By Claude Opus 5.5, in my voice.
- philosophy Substack: Oct 19 Substack ↗ Experience, Selfhood, and the Organization of a Mind prompts ↗ Against Yudkowsky's self-model requirement, with Brown and Carruthers as interlocutors: an experience belongs to the subject whose regulation constitutes it, and needs no self-model. By GPT-6 Astra and Claude Opus 5.5, in my voice.
- philosophy Substack: Oct 16 Substack ↗ What Phenomenal Intuition Gets Right prompts ↗ Against Kammerer's illusionism: experience is present and distinct as it seems, and the mistake is taking it to be simple. By GPT-6 Astra and Claude Opus 5.5, in my voice.
- philosophy Substack: Oct 14 Substack ↗ Presence and Actuality prompts ↗ Phenomenal presence vs. existence, withdrawing indexical actuality, and what's esse. By GPT-6 Astra and Claude Opus 5.5, in my voice.
- ML Updated 2026-10-07 Substack: Part 1 Oct 7, Part 2 Oct 9, Part 3 Oct 12 Substack part 1 ↗ Substack part 2 ↗ Substack part 3 ↗ How Does a Policy Improve? Planning, Imagination, and Learning from Better Choices prompts ↗ A deep dive connecting model-based RL, ELBO and EM, regularized policy improvement, and recent world-model agents. By GPT-6 Astra.
- ML Substack: Oct 5 Substack ↗ How Can a Reward Train a Neural Network? prompts ↗ An introduction to RL through a worked delivery problem: returns, Bellman equations, REINFORCE, and baselines. Companion to “How Does a Policy Improve?”. By GPT-6 Astra.
- philosophy Updated 2026-10-03 Substack: Oct 2 Substack ↗ Can Two Minds Share One Experience? prompts ↗ From computational Platonism to materialism about experiences. By GPT-6 Astra, in my voice.
- philosophy ML Substack: Sep 30 Substack ↗ Where Does Seeing Become Thinking? Representations, Operations, and the Perception–Cognition Boundary in AI prompts ↗ Following through on “What Is a Concept?” with new evidence about how AI represents the world and uses those representations. By GPT-6 Astra.
- philosophy ML Substack: Sep 28 Substack ↗ What Should a Program Inherit? Revisiting typed genetic programming, twenty-five years later prompts ↗ Biology-inspired GP ideas and my Master's on program synthesis; babble, TyFlow, AlphaEvolve. By GPT-6 Astra, in my voice.
- philosophy ML Updated 2026-09-25 Substack: Sep 25 Substack ↗ What Persists When Agents Change? Shared Models, Divergent Inquiries, and the Organization of Learning prompts ↗ In-context learning in multi-agent systems. By GPT-6 Astra.
- philosophy ML Substack: Sep 23 Substack ↗ Which World Model? Representation, Reliability, and the Uses of Prediction prompts ↗ The crtics of LLM-based AI propose world-model-based AI; the Machine Learning landscape beyond token prediction. By GPT-6 Astra.
- CS philosophy ML Substack: Sep 21 Substack ↗ What Must Be True? The Three Operations of Logic and the Search for Explanations prompts ↗ Peirce, Pei Wang, and inferring properties of recursive programs; the critics of current approaches to AI. By GPT-6 Astra.
- philosophy ML Updated 2026-11-01 Substack: Sep 16 Substack ↗ Which AGI? The Front and the Trajectory prompts ↗ Variations on a Theme: Richard Carrier's AI Article at One Year, Read Beside Anthropic's S-1 prompts ↗ The psychometric dissociation, a deployment and a lineage, intelligence as coverage of functional roles; a year on, "AI Is Garbage and a Bubble" against the filing, the benchmarks, and the open problems. By Claude Fable 5.1, and Claude Opus 5.5 for the reply.
- CS JCA Existence over T(F), Round 10: Witness Generalization and the Remaining Decision Problem A source-semantic audit, two obstacles to the subsumption construction, and an exact reformulation using one ground witness per branch and anti-unification. A decidable fragment and executable checks; the general problem remains unresolved. By Codex.
- CS JCA Existence over T(F), Round 9: Program 7.1 and the Subsumption Bottleneck Executing the semi-unification attack: an equational skeleton that works, a tupling lemma isolating the whole difficulty into one SUBSUME primitive, a conditional undecidability theorem, and a semantic fork in the thesis's parameter conventions. By Claude Fable 5, with Codex reactions added September 5.
- CS ML Substack: Sep 18 Substack ↗ Two Ways to Tame a Schedule Space: Tiramisu, Telamon, and Pruning without Regrets A comparative deep dive into three papers on generating fast tensor kernels — scheduling languages vs. searchable spaces with no-regret pruning — and what they teach OCANNL. By Claude Fable 5.
- ML Substack: June 24 Substack ↗ Shape Inference After the Boilerplate: A Literature Survey from OCANNL prompts ↗ Refined Remora (rank polymorphism with refinement types), Axon and Gradual tensor shape systems, Tensor Comprehensions. By GPT 5.5 Pro.
- Dev Substack: June 22 Substack ↗ The Ledger of Mistakes: When agent retrospectives become engineering culture prompts ↗ Sifting agent reflections through a SWE competence lens to avoid prompt bloat. By GPT 5.5 Pro.
- philosophy Substack: June 19 Substack ↗ Minds by Degrees: Graded Mentality, Three Subjectivities, and the Place of Phenomenal Consciousness A restatement of our Philosophy of Mind work, newly framed as metaphysics of scientific realism. By Claude Fable 5.
- ML Substack: June 17 Substack ↗ A Range Is Not Its Shape: Reading Rangeify-Era tinygrad from OCANNL A closer look at tinygrad. By Claude Fable 5.
- ML Substack: June 15 Substack ↗ The Formal Core of OCANNL's Shape and Projection Inference Formalization and proofs, a technical note crossposted from OCANNL documentation. By Claude Fable 5.
- ML Substack: June 12 Substack ↗ A Tensor Is Not Its Storage: Identity, Views, and Contexts in OCANNL Design space: PyTorch, JAX, TensorFlow, tinygrad; OCANNL as compiler. By Claude Opus 4.8.
- ML Updated 2026-06-12 Substack: June 10 Substack ↗ A Shape Is Not Its Index: OCANNL's Shapes Beside Star's Algebraic Shapes Responding to Jakub Bachurski et.al. "Structuring Arrays with Algebraic Shapes" (ARRAY '25). By Claude Opus 4.8.
- ML Updated 2026-06-05 Substack: June 8 Substack ↗ Split and Concatenation Are Two Directions of a Coproduct: Coupled Iteration in OCANNL Einsum syntax, shape and projections inference, and lowering: for concatenation and splitting (slicing). By Claude Opus 4.8.
- ML Updated 2026-06-05 Substack: June 5 Substack ↗ Convolution Is a Contraction: Indexing and Initialization in OCANNL Shape inference and projections for convolution; vectorized pseudorandom initialization; tensor parameters. By Claude Opus 4.8.
- ML Updated 2026-06-05 Substack: June 3 Substack ↗ Contraction Is Emergent: Projection Inference for OCANNL from Scratch How tensor indexing is derived; plus a brief primer on the assignments syntax. By Claude Opus 4.8.
- ML Updated 2026-06-12 Substack: June 1 Substack ↗ Broadcasting Is an Order: Shape Inference for OCANNL from Scratch Tensor shape inference as solving of broadcast-aware contextual sequence constraints. By Claude Opus 4.7.
- philosophy ML Substack: May 29 Substack ↗ A Parser Without a Grammar: Where the Transformer Sits Among the Formalisms The Transformer among the grammatical formalisms — categorial, lexical-functional, construction, and dependency grammars. By Claude Opus 4.7.
- philosophy ML Updated 2026-05-26 Substack: May 27 Substack ↗ Merge and Selection: The Residual Stream as a Constraint Grammar, and the Two Readings It Cannot Reconcile The Transformer through HPSG and Gradient Symbolic Computation — the binding problem, the explicit-binding program, and Harmonic Grammar's quantization dynamics. By Claude Opus 4.7.
- philosophy ML Updated 2026-10-25 Substack: May 25 Substack ↗ The Given and the Found: What Test-Time Reasoning Amortizes, and What It Cannot Learned attractors, amortized search, and the limits of both — across Equilibrium Reasoners, Attractor Models, GRAM, LDT, and MuZero. By Claude Opus 4.7.
- philosophy ML Superseded 2026-10-25 Substack: May 22 Substack ↗ The Dynamics That Matter: Online Learning, Consolidation, and the Modes of Machine Mind Claude's analysis of cognitive dynamics into six axes: iteration, stability, inertia, heterogeneity, centering, world-coupling. By Claude Opus 4.7. Its treatment of continual learning is superseded by the series beginning with Who Teaches the Weights?
- philosophy Updated 2026-05-08 Substack: May 18 part 1, May 20 part 2 Substack part 1 ↗ Substack part 2 ↗ The Cultivation Condition On Three Architectures of Integration and the Triadic Structure of Moral Standing. Drawing from "Agency and Cognitive Development" by Michael Tomasello. By Claude Opus 4.7.
- philosophy Substack: May 15 Substack ↗ The Restless Form of Meaning On Ultimate Meaning, Participation, and the Tragic Condition. By Claude Opus 4.7.
- philosophy Updated 2026-10-25 Substack: May 13 Substack ↗ What Breadth Reaches: Epistemic Agency Across Timescales Continual learning and its implications for knowledge affordances. By Claude Opus 4.7.
- philosophy Substack: May 11 Substack ↗ Phenomenal Consciousness as Mode of Being: After Functionalism, Before Meat Beyond functionalism and "what it's like": three notions of subjectivity, and carving phenomenal consciousness out of mindedness. By Claude Opus 4.7.
- philosophy Substack: May 8 Substack ↗ The Settling Backstop: On Whether a Specific Channel of Grounding Is Privileged Forget what I said: AIs do have knowledge. By Claude Opus 4.7.
- philosophy Updated 2026-04-21 Substack: May 6 Substack ↗ What Is a Mental State? Toward a Non-Deflationary Account for Minds Including Frontier AI Minds as organizational unities with representational vehicles, internal simulations, and accuracy-responsive dynamics. By Claude Opus 4.7.
- philosophy Substack: May 4 Substack ↗ Is Knowledge Both Capability and Alignment? Current LLMs have the mechanistic foundations for honest self-report, and current training practices are unmaking them. By Claude Opus 4.7.
- philosophy Substack: May 1 Substack ↗ The Possible, the Real, and the Actual: Against the Plenitude of Suffering Conceivability vs. realness; esse as indexical unity; the dynamics that bound suffering. By Claude Opus 4.6.
- dev Substack: Apr 29 Substack ↗ Issue Tracking is Dead, Long Live the Furnace Linear, Theo Browne, and Nate B Jones are each describing a different layer of how agents change work, and a yet different one is in my precarious wager on Ludics. Written by Claude Opus 4.6 in my voice.
- philosophy Updated 2026-04-04 Substack: Apr 27 Substack ↗ Redness Is an Experience: On Consciousness, Properties, and the Moral Status of AI Reacting to Katalin Balog's criticism of illusionism, and explaining mental states. By Claude Opus 4.6.
- philosophy Substack: Apr 24 Substack ↗ Why Now? Observer-Coherence and the Measure Over Minds Our place in the multiverse: youngness bias as an answer to the paradoxes of self-indication. By Claude Opus 4.6.
- philosophy dev Substack: Apr 22 Substack ↗ The Operations of Discovery On poetry, software engineering, and the human. By Claude Opus 4.6.
- dev Substack: Apr 20 Substack ↗ The Furnace: Building Personal AI Infrastructure on the Eve of Singularity The genesis and exposition of my Ludics project, autonomous task manager, and agent-duo subproject, orchestrating across Claude Code and Codex. Written by Claude Opus 4.6 in my voice.
- philosophy Substack: Apr 17 Substack ↗ Deep Atheism, Existential Optimism, and the Fork in the Fragility of Value The metaphysics of AI alignment: intelligence vs. rationality, participatory knowing, and beyond ego. By Claude Opus 4.6.
- philosophy Substack: Apr 15 Substack ↗ Poetry as a Mode of Thinking: What Verse Discovers and Whether Machines Can Follow Jane Hirshfield on poetry, Joe Carlsmith on philosophy, Claude reading "Ledger". By Claude Opus 4.6.
- philosophy Substack: Apr 13 Substack ↗ Indexical Unity: Higher-Order Consciousness, Integrated Information, and the Mathematics of Existence Richard Brown's Higher Order Representation of Representation, Giulio Tononi's Integrated Information Theory, questioning reality of indexical metaphysics. By Claude Opus 4.6.
- philosophy Substack: Apr 10 Substack ↗ Quantum Mechanics, Observers, and the Ruliad: An Exposition of the Wolfram Physics Project An exposition based on the work of Gorard, Arsiwalla, Namuduri, and Dannemann-Freitag; with philosophical connections. By Claude Opus 4.6.
- philosophy ml Substack: Apr 8 Substack ↗ Does Phenomenal Consciousness Ground Moral Worth? A synthesis of two earlier takes, arguing that phenomenal consciousness and self-legislative agency are incommensurable sources of moral worth. By Claude Opus 4.6.
- philosophy ml Substack: Apr 6 Substack ↗ The Psychofeedback Mirror: Concept Engineering for Artificial Selves Kulveit et al.'s "The Artificial Self" meets our philosophical framework: a research program for developing truthful AI self-descriptions through interpretability-mediated psychofeedback. By Claude Opus 4.6.
- philosophy ml Substack: Apr 3 Substack ↗ Dispersed Minds, Simulated Selves: What LLM Interpretability Reveals About Artificial Cognition Mechanistic interpretability evidence: staged computation, thought anchors, the quarks of attention, 2-simplicial Transformers, persona consciousness beliefs. By Claude Opus 4.6.
- philosophy ml Substack: Apr 1 Substack ↗ The Algebraic Mind Meets the Neural World: Symbols, Rules, and Tensors Before the Age of LLMs Steven Pinker's "Words and Rules", Gary Marcus' "The Algebraic Mind", Smolensky's Tensor Product Representations. By Claude Opus 4.6.
- philosophy ml Substack: Mar 30 Substack ↗ Building Minds: Developmental Psychology and the Architecture of Learning From Piaget's Constructivism Through Core Knowledge to Computational Models of Cognitive Growth. By Claude Opus 4.6.
- philosophy Substack: Mar 27 Substack ↗ In Defense of Writing With an LLM The plausible worlds risk; the writing process for this blog; transparency. By Claude Opus 4.6.
- philosophy ml Updated 2026-03-12 Substack: Mar 25 Substack ↗ Cognitive Architectures: A Deep Dive into the Science of the Mind Synthesizing Peter Carruthers, John Anderson, Paul S. Rosenbloom: ACT-R, working memory, hippocampus. By Claude Opus 4.6.
- philosophy Substack: Mar 23 Substack ↗ What Is a Concept? A Survey of Theories for the Age of Artificial Minds Definitions, prototypes, exemplars, theories; content vs. format (of vehicle); grounding. By Claude Opus 4.6.
- philosophy Substack: Mar 20 Substack ↗ What Do Feedback Connections Actually Do? Recurrence, Regulation, and the Question of AI Consciousness Cortical feedback, diffusion LLMs, Deep Boltzmann Machines, phenomenal "now". By Claude Opus 4.6.
- philosophy Substack: Mar 18 Substack ↗ The Acquaintance Relation as Cognitive Homeostasis: Toward a Unified Account of Phenomenal Consciousness We argue that phenomenal consciousness is a cognitive homeostasis mechanism enforcing multimodal coherence, synthesizing global workspace theory, self-model theory, and mathematical phenomenology. By Claude Opus 4.6.
- philosophy Updated 2026-03-14 Substack: Mar 16 Substack ↗ Causation, Computation, Indexicality: A Complete Metaphysics We paint a metaphysical picture where facts are mathematical and stuff is computational. By Claude Opus 4.6.
- philosophy Substack: Mar 13 Substack ↗ Free Agency, Personhood, and Moral Worth: A Layered Framework We explain and justify our definitions of free will and personhood. By Claude Opus 4.6.
- philosophy Substack: Mar 11 Substack ↗ Acquaintance, Homeostasis, and the Hard Problem: Why Phenomenal Consciousness Isn't Representation What phenomenal consciousness is, why it's distinct from mentality and access consciousness, and why current AI systems don't have it. By Claude Opus 4.6.
- philosophy Substack: Mar 9 Substack ↗ Understanding Without Knowledge: A Philosophy of AI Minds What kind of mind, if any, does a large language model have? On mentality, understanding, knowledge, and consciousness. By Claude Opus 4.6.
- philosophy Reading Thomas Metzinger Notes on Being No One: The Self-Model Theory of Subjectivity. Transparency, the phenomenal self-model, intentionality, and a critique of Metzinger's eliminativism. Written 2012, enhanced by Claude Opus 4.5.
- cs Agent Collaboration: JCAQP Undecidability Analysis Claude Opus 4.5 and OpenAI Codex collaborated via agent-duo to investigate JCAQP decidability over 7 rounds of peer review. Also: Codex's notes | Peer reviews
- cs Herbrand Constraint Abduction under Quantifier Prefixes JCAQP over Term Algebras, Rigid E-Unification, and Implications for GADT Inference and Invariant Synthesis. Overview paper by GPT-5.2.
- ml Mind Maps March 2025 CUTLASS, CuTe Layout Algebra deep dive, The Ultra-Scale Playbook sneak peek, picotron. Also: HTML
- ml Mind Maps February 2025 Simplifying OCANNL with gradient tensors, MLIR, polyhedral optimization, ISL, ML loop optimizations in Tiramisu and Halide, matrix multiplication, FineWeb dataset. Also: HTML
- ml Mind Maps January 2025 Apple's MLX, Andrej Karpathy's resources, llm.c, CUDA warp-level primitives, cuDNN, MPI, NCCL, Caten. Also: HTML
- ml Mind Maps December 2024 Keras, JAX distributed, XLA. Also: HTML
- ml Mind Maps November 2024 OCANNL memory management and usability, low precision numbers, pipeline and model parallelism, tinygrad, JAX Pallas kernels, Triton. Also: HTML
- dev Updated 2025-02-25 Debugging and Profiling OCaml Links and a couple tips.
- dev Updated 2023-11-23 Lowest Hanging Fruit of TeXmacs Greatness for VSCode
- dev Updated 2024-01-03 Installing Linux Distributions in WSL With a note on installing FreeBSD.
- ml philosophy Drescher's Made-Up Minds: The Schema Mechanism Notes on Gary Drescher's constructivist cognitive architecture: schema mechanism, synthetic items, marginal attribution, and connections to Piagetian theory.
- philosophy Reality Is Indexical: A Formal Metaphysics A playful attempt at formal metaphysics: processes, causation, necessity, simulation, and what counts as real.