DESK00:00:00:00
MAKER & MACHINE · THE PEOPLE WHO MAKE AI MATTER
VOL I · NO 001

Maker & MachinehAIstory · special section

hAIstory · The Race to AGI

How we got here

The Transformer (2017) was the inflection. OpenAI scaled it up. Anthropic broke off over how fast was too fast. The industry now competes for compute, electricity, and data — and most of the people building it think AGI lands between 2027 and 2032.

23
years
6
eras
103
milestones
46
labs & names

Latest from the wire

deal · Aug 17, 2026

NVIDIA / SB Energy / OpenAI. NVIDIA secured land, power, and shell at SB Energy's PORTS-Pike campus in Pike County, Ohio, to exclusively host NVIDIA AI compute. OpenAI is the tenant; NVIDIA is investing $1.5 billion in SB Energy. Read the article →

release · Aug 13, 2026

Google DeepMind. Google introduced Gemini 3.7 Flash, its workhorse model for coding and agents, at an introductory $0.75 / $3.75 per million input/output tokens through the end of 2026. Read the article →

release · Aug 11, 2026

Cursor / SpaceXAI. Grok Bot launched in beta: AI teammates with their own cloud computers, sold through Cursor Ultra and SuperGrok Heavy. Access widened August 21. Read the article →

Open the Now feed →

103 events shown. Press / to search, Esc to clear.

Era 1 · 1950 – 2011

Foundations

When AI was a dream, then a disappointment, then a dream again. Universities held the power. Funding came and went. The neural network was invented, dismissed, and quietly kept alive by a handful of believers.

Jan 5, 2020

Jan 5, 2020The Royal SocietyPeople of Science with Brian Cox - Dame Wendy Hall on Alan TuringID WW_ZFSCRuyY

Nov 6, 2013

Nov 6, 2013IBM ResearchWatson and the Jeopardy! ChallengeID P18EdAKuC1U

Era 2 · 2012 – 2017

The Deep Learning Revolution

Neural networks suddenly start working — not because of new ideas, but because GPUs got fast enough to run the old ones. Big Tech wakes up and starts buying.

Mar 13, 2020

Mar 13, 2020Google DeepMindAlphaGo - The Movie | Full award-winning documentaryID WXuK6gekU1Y

Jun 23, 2019

Jun 23, 2019Association for Computing Machinery (ACM)Geoffrey Hinton and Yann LeCun, 2018 ACM A.M. Turing Award Lecture "The Deep Learning Revolution"ID VsnQf7exv5I

Era 3 · 2018 – 2022

The Transformer & LLM Era

Transformer + huge data + huge GPUs = something nobody predicted. The first models that felt unsettlingly close to human language. Big Tech and OpenAI dominate. The big money starts moving.

Apr 8, 2024

Apr 8, 2024NVIDIA DeveloperTransforming AI | NVIDIA GTC 2024 Panel Hosted by Jensen HuangID hC_qASRcBhU

Nov 22, 2023

Nov 22, 2023Andrej Karpathy[1hr Talk] Intro to Large Language ModelsID zjkBMFhNj_g

Era 4 · Late 2022 – 2023

The ChatGPT Earthquake

AI stops being for researchers and becomes a household word. 100 million users in two months. Every CEO in the world calls an emergency meeting about AI strategy.

Mar 14, 2023

Mar 14, 2023OpenAIGPT-4 Developer LivestreamID outcGtbnMuQ

Nov 6, 2023

Nov 6, 2023OpenAIOpenAI DevDay: Opening KeynoteID U9mJuUkhUzk

Era 5 · 2018 – present

Two-Horse Race: OpenAI vs. Anthropic

Two labs, founded by people who used to work together, now define the frontier. The Microsoft–OpenAI vs. Amazon–Anthropic structural rivalry crystallizes. Every model launch is a counter-punch.

May 6, 2026

May 6, 2026Claude (Anthropic official)A conversation with Dario Amodei & Daniela AmodeiID 7xco5Qd2Oo8

Mar 18, 2024

Mar 18, 2024Lex FridmanSam Altman: OpenAI, GPT-5, Sora, Board Saga, Elon Musk, Ilya, Power & AGI | Lex Fridman Podcast #419ID jvqFAi7vkBc

Era 6 · 2023 – 2026

The Field Explodes

The race isn't two horses — it's a stampede. China enters the frontier. NVIDIA becomes the most valuable company on Earth. Electricity becomes the new oil.

May 13, 2024

May 13, 2024OpenAIIntroducing GPT-4oID DQacCB9tDaw

Sep 30, 2025

Sep 30, 2025OpenAIIntroducing Sora 2ID gzneGhpXwjU

Power Shifts

Who actually controlled AI — not just which model shipped.

1950 – 2005

Universities

AI lived in CS departments. MIT, Stanford, CMU, Toronto. Funded by DARPA, the NSF, and patient grants. The work was slow, public, and academic.

2010 – 2018

Big Tech

Google, Facebook, Microsoft, Baidu hire entire university labs. AlexNet wins ImageNet → industry sweeps in. DeepMind acquired. Brain, FAIR, MSR run the show. Universities become the farm system.

2018 – 2022

OpenAI + hyperscalers

GPT-2/3 prove scale wins. Microsoft's $1B locks OpenAI in. AWS, GCP, Azure become the only places frontier models can be trained. The cloud is now the lab.

2023 – 2026

GPU owners + frontier labs

NVIDIA at $5T. H100/Blackwell allocation is a strategic asset. OpenAI, Anthropic, Google DeepMind, xAI, DeepSeek own the frontier. Whoever has the GPUs and the talent runs the show.

The next 5 years

Energy + robotics infrastructure

When compute is commoditized and models are shared, the moat moves to electricity and physical-world embodiment. Whoever can power and embody AGI first defines the next industrial era. Nuclear deals, fab buildouts, and humanoid platforms become the new arms race.

OpenAI vs. Anthropic

WhenOpenAIAnthropic
2023-03GPT-4Claude 1
2023-07Claude 2 (public)
2023-11GPT-4 Turbo (128K context)Claude 2.1 (200K context)
2024-03Claude 3 (Haiku / Sonnet / Opus)
2024-05GPT-4o (text+image+audio+video)
2024-06Claude 3.5 Sonnet (beats Opus)
2024-09o1 — first reasoning model
2024-10Claude 3.5 Haiku + Computer Use
2024-11MCP open-sourced (industry standard)
2024-12o3 + Sora public
2025-02GPT-4.5 (Orion)Claude 3.7 + Claude Code
2025-05Claude 4 (Sonnet 4, Opus 4)
2025-08GPT-5
2025-09Claude 4.5 (first to break 80% on SWE-bench)
2025-11GPT-5.1 (Instant / Thinking)
2025-12GPT-5.2
2026-02GPT-5.3-CodexClaude 4.6 (Opus + Sonnet)
2026-04GPT-5.5Claude Opus 4.7
2026-06Fable 5

The Release Race

Click a lab to isolate its track.

2018

Jun GPT-1Oct BERT

2019

Feb GPT-2

2020

May GPT-3

2021

Aug Codex

2022

Apr PaLMNov ChatGPT (GPT-3.5)

2023

Feb LLaMAMar GPT-4Mar Claude 1Mar BardMay PaLM 2Jul Claude 2Jul Llama 2Sep Mistral 7BOct ERNIE 4.0Nov Grok-1Nov GPT-4 TurboNov Claude 2.1Dec Gemini 1.0Dec Mixtral 8x7B

2024

Feb Gemini 1.5Mar Claude 3Apr Llama 3May GPT-4oJun Claude 3.5 SonnetJul Llama 3.1 405BAug Grok-2Sep o1Sep Qwen 2.5Oct Claude 3.5 Haiku + Computer UseDec Gemini 2.0Dec o3 (announced)Dec DeepSeek V3

2025

Jan DeepSeek R1Feb Grok 3Feb Claude 3.7 SonnetFeb GPT-4.5Mar Gemini 2.5Apr Llama 4Apr o3 + o4-miniApr Qwen 3May Claude 4Jul Grok 4Jul Kimi K2Aug gpt-ossAug Claude Opus 4.1Aug GPT-5Sep Claude Sonnet 4.5Oct Claude Haiku 4.5Nov GPT-5.1Nov Gemini 3Nov Claude Opus 4.5Dec GPT-5.2

2026

Feb GPT-5.3-CodexFeb Claude Opus 4.6Feb Claude Sonnet 4.6Apr Claude Opus 4.7Apr GPT-5.5Jun Claude Fable 5Apr DeepSeek-V4 PreviewJun Claude Sonnet 5Aug Gemini 3.7 Flash

What the builders predict

CEO, OpenAI

Sam Altman

  • 2025. AI agents "join the workforce." Read the article →
  • 2026. Systems "discover novel insights."
  • 2027. Real-world humanoid robots.
  • 2032–35. Superintelligence ("a few thousand days"). Read the article →

CEO, Anthropic

Dario Amodei

  • Oct 2024. "Machines of Loving Grace" essay published. Read the article →
  • 2026. Powerful AI plausibly arrives.
  • 2030–35. "Compressed 21st century" — 50–100 years of biomedical progress in 5–10. Read the article →

CEO, Google DeepMind

Demis Hassabis

  • 2025. States AGI is 5–10 years out.
  • 2030–35. AGI plausibly arrives.
  • Beyond. "10× the Industrial Revolution, happening 10× faster."

2025–26

Agentic AI in production. Reasoning models become standard.

2026–27

First serious AI scientific discoveries. Real coding autonomy.

2027–30

Plausible window for AGI (general human-level AI).

2030–35

Plausible window for superintelligence (AI beyond humans).

The real constraints

Electricity

Nuclear restarts, grid buildout. May be the real bottleneck.

Chip supply

NVIDIA, TSMC, geopolitics around Taiwan.

Data

The "data wall" — synthetic data is the bet.

Capital

$100B+ training runs, trillion-dollar infrastructure.

Regulation

EU AI Act, US executive orders, China policy.

A few things worth remembering

  • The Transformer (2017) is the technical break. Everything since is scaling, refining, deploying that idea.
  • OpenAI and Anthropic share a common ancestor. Anthropic exists because some of OpenAI's people thought the lab was moving too fast.
  • Open-source is now competitive. LLaMA and DeepSeek proved frontier quality does not have to be locked up.
  • The bottleneck is physical. Electricity, GPUs, data — not a shortage of papers.
  • The public AGI window is 2027–2032, with a wide private interval even among the optimists.

Primary sources

Foundational papers

Modern model papers

Recent models & releases

Money, compute & litigation