The research papers behind Anthropic

Before they started Anthropic, its founders co-wrote some of the most influential AI research papers of 2016 to 2020 at OpenAI and Google, including the paper that introduced GPT-3, an early forerunner of ChatGPT. Autumn listed every author of ten of those papers, in order, and checked who later joined Anthropic.

10 of 31

authors of the GPT-3 paper later joined Anthropic

6 of 10

authors of the scaling-laws paper did too

future Anthropic people across ten research papers

authors checked

The GPT-3 paper and its 31 authors

The research paper behind OpenAI's 2020 GPT-3 model, Language Models are Few-Shot Learners, lists 31 authors, shown here in the paper's order. Ten later joined Anthropic, six of them as co-founders, including the first two authors. For the rest, each card shows their current job where a source states it (checked October 2026).

Ten research papers, 2016 to 2020

Newest first. Each card says what the paper showed, in plain words, and who among its authors later joined Anthropic. Each square is one author, in the paper's order; orange marks someone who later worked there.

2020 · GPT-310 of 31 authors later joined

Language Models are Few-Shot Learners

Introduced GPT-3, a 175-billion-parameter language model that could take on new tasks from just a few examples written in the prompt.

Tom Brown, later at AnthropicBen Mann, later at AnthropicNick RyderMelanie SubbiahJared Kaplan, later at AnthropicPrafulla DhariwalArvind NeelakantanPranav ShyamGirish SastryAmanda Askell, later at AnthropicSandhini AgarwalAriel Herbert-VossGretchen KruegerTom Henighan, later at AnthropicRewon ChildAditya RameshDaniel Ziegler, later at AnthropicJeffrey Wu, later at AnthropicClemens WinterChristopher HesseMark ChenEric SiglerMateusz LitwinScott GrayBenjamin ChessJack Clark, later at AnthropicChristopher BernerSam McCandlish, later at AnthropicAlec RadfordIlya SutskeverDario Amodei, later at Anthropic

20208 of 19 authors later joined

Scaling Laws for Autoregressive Generative Modeling

Showed that the scaling laws found for language models also hold for models of images, video, math, and paired images and text.

Tom Henighan, later at AnthropicJared Kaplan, later at AnthropicMor KatzMark ChenChristopher HesseJacob JacksonHeewoo JunTom Brown, later at AnthropicPrafulla DhariwalScott GrayChris HallacyBen Mann, later at AnthropicAlec RadfordAditya RameshNick RyderDaniel Ziegler, later at AnthropicJohn Schulman, later at AnthropicDario Amodei, later at AnthropicSam McCandlish, later at Anthropic

2020 · Scaling laws6 of 10 authors later joined

Scaling Laws for Neural Language Models

Found that language models improve in a predictable way as they get bigger and are trained with more data and computing power.

Jared Kaplan, later at AnthropicSam McCandlish, later at AnthropicTom Henighan, later at AnthropicTom Brown, later at AnthropicBenjamin ChessRewon ChildScott GrayAlec RadfordJeffrey Wu, later at AnthropicDario Amodei, later at Anthropic

2020 · Summarize from feedback3 of 9 authors later joined

Learning to Summarize from Human Feedback

Trained models to summarize text using human feedback; people preferred their summaries to the human-written reference summaries.

Nisan StiennonLong OuyangJeffrey Wu, later at AnthropicDaniel Ziegler, later at AnthropicRyan LoweChelsea VossAlec RadfordDario Amodei, later at AnthropicPaul Christiano

2020 · Zoom In2 of 6 authors later joined

Zoom In: An Introduction to Circuits

Argued that a neural network can be understood piece by piece, as circuits of features a person can interpret.

Chris Olah, later at AnthropicNick CammarataLudwig SchubertGabriel GohMichael PetrovShan Carter, later at Anthropic

2019 · Fine-tuning from preferences4 of 8 authors later joined

Fine-Tuning Language Models from Human Preferences

Applied learning from human preferences to language models, teaching them to continue and summarize text the way people preferred.

Daniel Ziegler, later at AnthropicNisan StiennonJeffrey Wu, later at AnthropicTom Brown, later at AnthropicAlec RadfordDario Amodei, later at AnthropicPaul ChristianoGeoffrey Irving

2019 · GPT-22 of 6 authors later joined

Language Models are Unsupervised Multitask Learners

Showed that a 1.5-billion-parameter language model could translate, summarize and answer questions without being trained for those tasks.

Alec RadfordJeffrey Wu, later at AnthropicRewon ChildDavid LuanDario Amodei, later at AnthropicIlya Sutskever

2018 · Building Blocks2 of 7 authors later joined

The Building Blocks of Interpretability

Built visual tools for seeing what an image-recognition model has learned and how it reaches its answers.

Chris Olah, later at AnthropicArvind SatyanarayanIan JohnsonShan Carter, later at AnthropicLudwig SchubertKatherine YeAlexander Mordvintsev

2017 · RL from human preferences3 of 6 authors later joined

Deep Reinforcement Learning from Human Preferences

Trained AI agents from people's choices between pairs of short clips instead of a hand-written goal, the approach later called reinforcement learning from human feedback (RLHF).

Paul ChristianoJan Leike, later at AnthropicTom Brown, later at AnthropicMiljan MarticShane LeggDario Amodei, later at Anthropic

2016 · Concrete Problems3 of 6 authors later joined

Concrete Problems in AI Safety

Set out five practical problems in making AI systems behave safely, such as gaming the goal they are given or causing unwanted side effects.

Dario Amodei, later at AnthropicChris Olah, later at AnthropicJacob SteinhardtPaul ChristianoJohn Schulman, later at AnthropicDan Mané

Sources¹arxiv.org ²arxiv.org ³distill.pub ⁴arxiv.org ⁵cdn.openai.com ⁶arxiv.org ⁷arxiv.org ⁸arxiv.org ⁹arxiv.org ¹⁰distill.pub

Which future Anthropic people wrote which paper

The 14 authors who later joined Anthropic, and the research papers each one co-wrote. Nine have a profile on these pages.

Author	Concrete Problems · 2016	RL from human preferences · 2017	Building Blocks · 2018	Fine-tuning from preferences · 2019	GPT-2 · 2019	GPT-3 · 2020	Summarize from feedback · 2020	Scaling laws (images, video) · 2020	Scaling laws · 2020	Zoom In · 2020	Joined
Dario Amodeico-founder	author	author	no	author	author	author	author	author	author	no	2021
Tom Brownco-founder	no	author	no	author	no	author	no	author	author	no	2021
Chris Olahco-founder	author	no	author	no	no	no	no	no	no	author	2021
Jared Kaplanco-founder	no	no	no	no	no	author	no	author	author	no	2021
Sam McCandlishco-founder	no	no	no	no	no	author	no	author	author	no	2021
Ben Mannco-founder	no	no	no	no	no	author	no	author	no	no	2021
Jack Clarkco-founder	no	no	no	no	no	author	no	no	no	no	2021
Jeffrey Wu	no	no	no	author	author	author	author	no	author	no	2024 · left
Daniel Ziegler	no	no	no	author	no	author	author	author	no	no	2024
Tom Henighan	no	no	no	no	no	author	no	author	author	no	2021
John Schulman	author	no	no	no	no	no	no	author	no	no	2024 · left
Shan Carter	no	no	author	no	no	no	no	no	no	author	2022
Amanda Askell	no	no	no	no	no	author	no	no	no	no	2021
Jan Leike	no	author	no	no	no	no	no	no	no	no	2024

More on Anthropic

The peopleAnthropic's leadership team and org chartAn org chart of the co-founders, the executive team and the researchers behind Claude, with a profile for each.Open →The talentWhere Anthropic's people came fromThe OpenAI founding wave, the later DeepMind wave, the Stripe pipeline into leadership and the YC founders who joined.Open →The moneyAnthropic IPO, valuation and revenueEvery funding round, the revenue run-rate, who owns Anthropic and where the IPO stands.Open →

How we made this

Research task Completed

Prompt

Map the authors of the research papers that the founders of Anthropic (the AI company that makes Claude, founded 2021) wrote together before Anthropic existed, and where each author went. Papers (use the official arXiv or Distill author lists): 1. Concrete Problems in AI Safety (2016)… 8. Language Models are Few-Shot Learners (GPT-3, 2020)…

Papers: 10
Author slots: 108
People: 58