THE AI MUSEUM

Nine years that changed everything. Walk through them.

Admission

The guided tour

Show me the essentials

A curated route through the works that changed the direction of the field.

8 works/about 15 minutes

Free roam

Let me walk the building

Every work, in order, through all five rooms of the long gallery.

31 works/5 rooms

Plan of the building

8 rooms
THE LONG GALLERY01Foundation2017–2019 · 302Scale2020–2022 · 503Inflection2022–2023 · 704Intelligence2024–2025 · 705Agentic2025–2026 · 9Pioneers6 PORTRAITSArtifacts5 OBJECTSHorizon3 REFLECTIONS

Select a room to enter, or simply walk through.

The floor line

The scale of it

Every parameter count the field has published, on a logarithmic scale. Each gridline is ten times the one below it. The rise is the part everyone expects. The gaps are the part worth standing in front of.

65M → 2.8T/4.6 orders

100M1B10B100B1T10T20172019202120232025202615 MILESTONES PUBLISHED NO FIGURETransformer65MGPT-3175BLLaMA65BDeepSeek R1671BKimi K32.8T
Open weights (filled)Closed, figure published (hollow)No figure published (tick)

Every parameter count published since 2023 belongs to a model whose weights you can download. Not one of the closed frontier systems of the last three years has stated its size. The measure did not get harder to take. It stopped being offered.

View as table
Disclosed parameter counts by model and year
ModelYearParametersWeights
Transformer201765MOpen
BERT2018340MOpen
GPT-220192BOpen
GPT-32020175BClosed
DALL·E202112BClosed
Codex202112BClosed
DALL·E 220224BClosed
Stable Diffusion2022890MOpen
LLaMA202365BOpen
Llama 2202370BOpen
Mistral 7B20237BOpen
Llama 3202470BOpen
DeepSeek R12025671BOpen
Llama 42025109BOpen
DeepSeek V420261.6TOpen
Kimi K320262.8TOpen
Wing III
Framed portraits hung on a warm gallery wall
Wing III — The Pioneers Room

The Pioneers Room

The minds that made it.

Six people whose work — across seven decades — created the conditions for everything that followed.

Alan Turing

Theorist

Alan Turing

Can machines think? The question he posed in 1950 defined the terms of a debate that still runs today. Everything that followed traces back to that single sentence.

John McCarthy

Founder

John McCarthy

He coined the term Artificial Intelligence. The Dartmouth conference he organised gave a discipline its name — and with a name, a direction.

Geoffrey Hinton

Architect

Geoffrey Hinton

Backpropagation made deep networks trainable. The algorithm he helped establish is still the engine behind every model built today.

Yann LeCun

Visionary

Yann LeCun

Convolutional networks taught machines to see. His architecture from the late 1980s became the foundation of modern computer vision.

Ilya Sutskever

Catalyst

Ilya Sutskever

AlexNet broke ImageNet. The paper he co-authored at 26 triggered the deep learning wave that reshaped the entire industry within a decade.

Andrej Karpathy

Translator

Andrej Karpathy

Made the frontier legible. His writing and teaching turned obscure research into practical intuition for an entire generation of engineers.

Wing IV
A lit vitrine holding a single catalogued object
Wing IV — The Artifacts Vault

The Artifacts Vault

The objects that changed everything.

Five papers, datasets, and moments whose consequences are still unfolding.

2017
Artifact · 001Research Paper

Attention Is All You Need

Vaswani, Shazeer, Parmar et al. · Google Brain

Fifteen pages. Eight authors. The Transformer architecture they introduced powers every major language model built since. No recurrence, no convolution — just attention. The most cited paper in the history of machine learning.

2009
Artifact · 002Dataset

The ImageNet Dataset

Fei-Fei Li et al. · Stanford Vision Lab

1.2 million labeled images across 1,000 categories. The dataset that gave deep learning its first real test. Without ImageNet, the 2012 AlexNet moment — the moment the field turned — never happens.

2020
Artifact · 003Research Paper

Scaling Laws for Neural Language Models

Kaplan, McCandlish, Henighan et al. · OpenAI

Bigger is better — and predictably so. This paper showed that model performance scales as a power law with compute, data, and parameters. The insight that justified billions in investment and set the agenda for the frontier.

2022
Artifact · 004Technical Report

Constitutional AI

Bai, Jones, Ndousse et al. · Anthropic

Safety as architecture, not afterthought. Anthropic published a method for aligning AI systems using a written constitution — a set of principles — rather than exhaustive human feedback alone.

2022
Artifact · 005Industry Event

The ChatGPT Launch

OpenAI · November 30, 2022

One hundred million users in sixty days. No product in history had grown that fast. ChatGPT didn't invent large language models — it made them real to everyone. Nothing about the conversation was the same after.

Wing V
A bright, empty gallery receding under skylights
Wing V — The Horizon

Horizon

Where does it go from here?

Every museum ends with a door back to the world outside. This is that door — with three things worth carrying through it.

The question

Alan Turing asked in 1950 whether machines could think. Seventy-six years later, the question hasn't been answered — it's been reframed. We stopped asking whether machines could think and started asking whether it matters.

The pattern

Every era in this museum followed the same arc: a paper no one believed, a model that shouldn't have worked, a moment the world changed overnight. The arc is still running.

The open door

The years ahead have no exhibit yet. No artifact has been catalogued. No pioneer has been named. That wing is still being built — and the people building it are alive right now.

End of exhibition

Return to the entrance

Picture credits

Photography via Pexels

Interior photography is used to give each room its character. It is illustrative rather than documentary: these are not photographs of the works described.