# LLM Bento

> A curated guide to understanding AI language models — their capabilities, stories, and the questions they raise.

Append `.md` to any page path to get its markdown content.

## Learn

- [The Wonder](/learn/the-wonder.md): How can a computer speak like a human?
- [The Core Insight](/learn/prediction.md): Next token prediction — the mechanism behind it all
- [Different Minds](/learn/human-vs-llm.md): Human vs LLM understanding
- [The Birth](/learn/birth-of-llm.md): How LLMs come to life through training
- [The Mystery](/learn/the-mystery.md): What we don't fully understand about AI
- [Beyond Text](/learn/multimodal.md): What a token really is, and how one architecture learned to see, hear, and create

## Thought Experiments

- [The Turing Test](/thought-experiments/turing-test.md): Can machines think? (Turing, 1950)
- [The Chinese Room](/thought-experiments/chinese-room.md): Does symbol manipulation equal understanding? (Searle, 1980)
- [Mary's Room](/thought-experiments/marys-room.md): Can physical knowledge capture subjective experience? (Jackson, 1982)
- [What Is It Like to Be a Bat?](/thought-experiments/nagels-bat.md): The limits of understanding other minds (Nagel, 1974)
- [The Philosophical Zombie](/thought-experiments/philosophical-zombie.md): Behavior without experience? (Chalmers, 1996)
- [The Brainstorm Machine](/thought-experiments/brainstorm-machine.md): Could we ever share experience? (Dennett, 1991)
- [Grokking](/thought-experiments/grokking.md): When does learning become understanding?
- [The Day-Man and the Night-Man](/thought-experiments/day-night-man.md): One body, two people? Personal identity for LLM threads (Locke/Chalmers, 1690/2026)
- [Boltzmann Brains](/thought-experiments/boltzmann-brains.md): Can a mind trust its own reasoning? Cognitive instability and where beliefs come from (Boltzmann/Carroll, 1896/2017)

## Thinking Tools

- [Thinking Tools](/thinking-tools.md): Curated instruments for examining minds — thought experiments, intuition pumps, and frameworks
- [Intuition Pumps](/thinking-tools/intuition-pumps.md): Dennett's technique for philosophical investigation — find the hidden knobs and turn them

## Essays

- [Agency](/agency.md): The capacity to shape reality through deliberate action
- [AI Alignment](/alignment.md): Why getting AI to do what we want is harder than it sounds
- [Four Kinds of Knowing](/knowings.md): Vervaeke's framework for human understanding
- [AI Moments](/moments.md): A timeline of pivotal events in AI history (2016–2025)

## Story

- [The LLM Story](/story.md): An interactive narrative about how we got here

## Threads

- [Threads](/threads.md): Slow news — stories that do not finish, tracked as the positions in a public argument and who holds each. Currently one thread: [Should the frontier slow down](/threads/pacing-the-frontier.md), with five positions from Amodei and Pachocki to Huang and Suleyman. Traditional Chinese: /zh-tw/threads.md

## Explore

- [Explore](/explore.md): A curated guide to understanding AI — videos, books, podcasts, and essays following a natural curiosity arc

## Questions

- [The Question Streams](/questions.md): The open questions the whole site orbits — trunk questions and their descendants, mapped as lineage rivers through time
- [Where does this argument live?](/questions/layers.md): Five layers an argument about LLMs can stand on — society, harness, post-training, pre-training, knowing — and three arguments made on one layer that can only be settled on another
- [What is understanding?](/questions/what-is-understanding.md): One node from the streams — the cuts of Searle, Hinton, and Bender
- [What is real?](/questions/what-is-real.md): Plato's Forms, Kant's mind-structured reality, Vervaeke's participatory realness, Chalmers' virtual realism
- [What is technology?](/questions/what-is-technology.md): Aristotle's techne, Heidegger's enframing, McLuhan's medium as message
- [What is beauty?](/questions/what-is-beauty.md): Plato's Form of beauty, Kant's disinterested judgment, Gadamer's beauty as truth
- [What is art?](/questions/what-is-art.md): Tolstoy's transmission of feeling vs. Danto's artworld
- [How does a mind know what matters?](/questions/how-does-a-mind-know-what-matters.md): Vervaeke's relevance realization as the standing answer to Dennett's frame problem, with Friston's salience and Levin's scale-indexed concern
- [What is a self?](/questions/what-is-a-self.md): Varela's virtual, selfless self; Bach's self-model; Levin's light-cone boundary; Vervaeke's agent-arena relationship
- [Are we separate from our machines?](/questions/are-we-separate-from-our-machines.md): McLuhan's extensions of man, Clark's natural-born cyborgs, Agüera y Arcas's human–machine symbiosis

## Channels

- [Channels](/channels.md): The channels this video collection keeps drawing on for long-form conversations — eleven hand-picked sources with what each one is for, and the full long tail of 165 other channels that feed the collection. Each has its own page: /channels/lex-fridman.md, /channels/dwarkesh-patel.md, /channels/eighty-thousand-hours.md, /channels/big-think.md, /channels/theories-of-everything.md, /channels/machine-learning-street-talk.md, /channels/welch-labs.md, /channels/startalk.md, /channels/world-science-festival.md, /channels/palisade-research.md, /channels/hannah-fry.md — with that channel's full roster of people and every conversation drawn from it, except /channels/hannah-fry.md, which lists her appearances recorded elsewhere because nothing in the collection comes from her own channel. Traditional Chinese: /zh-tw/channels.md

## People & Connections

- [Connections Graph](/api/graph.json): The full people-connections graph as JSON — 130+ people and 110+ authored edges between them. Every edge has a type (influenced-by, collaborated, conversed, debated, kindred, contrast), a curatorial weight (0–100), a written "why", and sources for factual claims. This is an editorial graph, written and fetch-verified by hand — not scraped. Start here for the site's map of who shaped whom.
- Person pages as markdown: `/people/{id}.md` (e.g. `/people/richard-sutton.md`) — bio, contributions, curated connections with reasons, videos, and links. Traditional Chinese: `/zh-tw/people/{id}.md`. Person IDs are in the graph JSON.

## Optional

- [Models](/models): Browse 35+ AI models across 7 companies
- [Compare](/compare): Side-by-side model comparison
- [Concepts](/concepts): Quick reference glossary