Library

Books, papers, and thinkers that shaped how I think about intelligence, mathematics, and cognition. Not a reading list — an intellectual bibliography. These are the works I return to, argue with, and build on.

Canon

Books I return to. Each one changed something fundamental in how I see a problem that still matters.

Gödel, Escher, Bach: An Eternal Golden Braid
Douglas Hofstadter · 1979
The most serious attempt ever written to explain consciousness through self-reference and strange loops. Required reading before building any AI system that claims to "understand" anything. Still unsurpassed.
The Emperor's New Mind
Roger Penrose · 1989
A physicist's argument that consciousness requires something beyond computation. Wrong or right, it asks the question correctly — and no serious response has fully disposed of it.
The Feynman Lectures on Physics
Richard Feynman · 1964
Not physics textbooks. A demonstration that deep understanding has a distinct texture you can recognise and aim for. Clarity as a form of rigor, not an alternative to it.
Probability Theory: The Logic of Science
E.T. Jaynes · 2003
Bayesian inference derived from first principles as an extension of classical logic. The book that makes probability feel inevitable rather than arbitrary. Foundation for how I think about uncertainty in AI systems.
Thinking, Fast and Slow
Daniel Kahneman · 2011
The rigorous empirical demolition of the rational agent model. Any AI system that interacts with humans needs dual process theory as a foundation. The gap between System 1 and System 2 is an engineering problem.
The Society of Mind
Marvin Minsky · 1986
The original multi-agent system: intelligence as the emergent behavior of many small, dumb processes with no central controller. Four decades before transformers, Minsky got the architecture right.
The Structure of Scientific Revolutions
Thomas Kuhn · 1962
Required reading for anyone building AI systems who thinks they are doing science. Most AI research is normal science; occasionally something shifts the paradigm. Recognising which is which is a skill.
An Introduction to Information Theory
John R. Pierce · 1961
Shannon's framework, explained accessibly. Information is physical. Entropy is not a metaphor. The mathematical foundation underneath every loss function ever written.

Key Papers

Papers that changed how I think about specific problems. One sentence on why each one matters.

Agents
ReAct: Synergizing Reasoning and Acting in Language Models
Yao et al. · 2022
arXiv →
Toolformer: Language Models Can Teach Themselves to Use Tools
Schick et al. · 2023
arXiv →
A Survey on Large Language Model based Autonomous Agents
Wang et al. · 2023
arXiv →
MemGPT: Towards LLMs as Operating Systems
Packer et al. · 2023
arXiv →
Models
Attention Is All You Need
Vaswani et al. · 2017
arXiv →
Scaling Laws for Neural Language Models
Kaplan et al. · 2020
arXiv →
Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
Wei et al. · 2022
arXiv →
Training language models to follow instructions with human feedback
Ouyang et al. · 2022 (InstructGPT)
arXiv →
Mathematics
A Mathematical Theory of Communication
Claude Shannon · 1948
PDF →
On the Free Energy Principle
Karl Friston · 2010
Nature →
Physical AI
World Models
Ha & Schmidhuber · 2018
arXiv →
RT-2: Vision-Language-Action Models
Brohan et al. · 2023
arXiv →

Thinkers

Not a hall of fame. Honest notes on why each person's thinking changed how I see a specific problem.

Ramanujan Mathematical truth can be intuited before it can be proved. The counterexample to the idea that rigor and understanding are the same thing. What he achieved is a standing challenge to every purely computational view of cognition.
Leibniz The original vision: reasoning as mechanical symbol manipulation. The idea that intelligence could be formalised is older than computers by three centuries. Leibniz was the first AI researcher.
Euclid For the demonstration that an entire edifice of knowledge can be derived from a small number of axioms. The model for what it means to understand something from first principles.
Gauss For the principle that mathematical elegance and empirical accuracy are not in tension — they converge. The normal distribution is both beautiful and everywhere.
Gödel On the hard limits of formal systems. Every AI architecture that claims to reason is implicitly navigating what Gödel showed cannot be navigated from within. Understanding incompleteness is not optional.
von Neumann Architecture as epistemology. How you structure computation shapes what you can compute. The von Neumann bottleneck is not just an engineering problem — it's a cognitive one.
Shannon Information is physical. Entropy is not a metaphor. Every loss function, every compression algorithm, every communication protocol is a consequence of Shannon's 1948 paper.
Turing The original definition. What computation means — and, crucially, what it cannot mean. The halting problem is the oldest alignment problem.
Feynman For the principle that you don't understand something until you can derive it from scratch. And for the demonstration that clarity is a form of rigor, not an alternative to it.
Penrose The most serious attempt to connect physics to consciousness, taken seriously by someone who knows both. Wrong or right, The Emperor's New Mind asks the question in a way that hasn't been answered.
Kahneman The empirical demolition of the rational agent. Any AI system that interacts with humans needs dual process theory as a foundation. The gap between System 1 and System 2 is an engineering challenge.
Minsky Intelligence as the emergent behavior of many small, stupid processes. The original multi-agent system. Decades before transformers, Minsky understood that mind might not have a center.
Hofstadter Strange loops as the key to self-reference, consciousness, and meaning. GEB remains the most honest book written about what it would even mean for a formal system to understand itself.
Sri Aurobindo On consciousness as ontologically primary, not emergent from matter. A necessary counterweight to purely computational views of mind. The question of whether intelligence requires consciousness is not settled.
Carl Sagan For the cosmic perspective. Intelligence is rare in the universe — we know of exactly one planet that has it. The stakes of getting AI right are not just economic.