DACHVARD

~/library ~/writing ~/author ~/wander

◆ ◇ ◆

BIBLIOTHECA

The Reading Ledger

A log of books and papers, current and past.

— ◆ —

Currently Reading

The Book of Disquiet — Fernando PessoaThird reading. New rooms appear each time.	reading
Mechanistic Interpretability of Neural Networks — Anthropic (Elhage et al.)Superposition, circuits, the slow archaeology of residual streams.	reading
Gödel, Escher, Bach — Douglas HofstadterRe-reading with agent architectures in mind. The loops feel different now.	reading

Read This Year

Attention Is All You Need — Vaswani et al.Still the most important eight pages in the field.	complete
Chain-of-Thought Prompting Elicits Reasoning in Large Language Models — Wei et al.The paper that made everyone think reasoning was solved. It was not.	complete
Invisible Cities — Italo CalvinoA model for how to describe things that do not exist but should.	complete
Proofs and Refutations — Imre LakatosHow mathematics actually develops. Required reading for anyone building systems.	complete
Constitutional AI: Harmlessness from AI Feedback — AnthropicThe alignment approach I find most honest about its own assumptions.	complete
The Name of the Rose — Umberto EcoA novel about libraries, signs, and what happens when you mistake the index for the book.	complete
Scaling Laws for Neural Language Models — Kaplan et al.The empirical foundation. The exponents still puzzle me.	complete
Tractatus Logico-Philosophicus — Ludwig WittgensteinThe limits of language, stated at the limits of language.	complete

The Pile

The Phenomenology of Spirit — G.W.F. HegelPromised myself I would read it properly this year.	in pile
Tree of Thoughts: Deliberate Problem Solving with LLMs — Yao et al.On the reading list since it came out.	in pile
The Myth of Sisyphus — Albert CamusRelevant to training runs.	in pile
Information Theory, Inference, and Learning Algorithms — David MacKayThe canonical text. I keep starting it and getting lost in footnotes.	in pile
Ficciones — Jorge Luis BorgesTo re-read whenever I forget why any of this matters.	in pile
Reward is Enough — Silver, Singh, et al.The maximalist claim. I am skeptical. I should read it carefully.	in pile

❦

✦ memory · ☽ night · ∞ loops · ❧ margins · ◆ proof

a personal library in perpetual arrangement · MMXXVI