# The research papers behind Anthropic

Every author of ten AI research papers Anthropic's founders co-wrote before the company existed, including the GPT-3 paper, 2016 to 2020, and which of them later joined Anthropic. Researched with Autumn (https://www.autumn.ai) from each paper's own author list and public career sources. Updated 2026-10-18. Interactive version: https://www.autumn.ai/resources/anthropic/papers

## Language Models are Few-Shot Learners (2020)

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

10 of 31 authors later joined Anthropic. Paper: https://arxiv.org/abs/2005.14165

1. [Tom Brown](https://www.autumn.ai/resources/anthropic/people/tom-brown) (Anthropic co-founder) [1](https://nextomoro.com/tom-brown/) [2](https://www.linkedin.com/in/nottombrown)
2. [Ben Mann](https://www.autumn.ai/resources/anthropic/people/ben-mann) (Anthropic co-founder) [1](https://www.linkedin.com/in/benjamin-mann)
3. [Nick Ryder](https://arxiv.org/search/?searchtype=author&query=Ryder%2C+Nick); now OpenAI, VP of Research, Foundations [1](https://naturalsciences.rice.edu/news/natural-sciences-names-collier-59-ryder-14-2026-alumni-award-recipients)
4. [Melanie Subbiah](https://melaniesubbiah.github.io/); now Brown University, postdoctoral researcher [1](https://melaniesubbiah.github.io/)
5. [Jared Kaplan](https://www.autumn.ai/resources/anthropic/people/jared-kaplan) (Anthropic co-founder) [1](https://www.hertzfoundation.org/people/jared-kaplan/) [2](https://www.linkedin.com/in/jared-kaplan-645843213)
6. [Prafulla Dhariwal](https://www.linkedin.com/in/prafulladhariwal); now OpenAI, technical fellow [1](https://www.linkedin.com/in/prafulladhariwal)
7. [Arvind Neelakantan](https://arvind-neural.github.io/); now Google DeepMind, research scientist [1](https://arvind-neural.github.io/)
8. [Pranav Shyam](https://arxiv.org/search/?searchtype=author&query=Shyam%2C+Pranav); now Google DeepMind, research scientist [1](https://www.alphaxiv.org/organizations/openai/alumni)
9. [Girish Sastry](https://www.linkedin.com/in/girish-sastry-2a39348); now Independent, AI policy research and advising [1](https://www.linkedin.com/in/girish-sastry-2a39348)
10. [Amanda Askell](https://www.autumn.ai/resources/anthropic/people/amanda-askell) (joined Anthropic 2021) [1](https://en.wikipedia.org/wiki/Amanda_Askell) [2](https://www.linkedin.com/in/amanda-askell)
11. [Sandhini Agarwal](https://arxiv.org/search/?searchtype=author&query=Agarwal%2C+Sandhini)
12. [Ariel Herbert-Voss](https://www.linkedin.com/in/adversariel); now RunSybil, co-founder and CEO [1](https://www.linkedin.com/in/adversariel)
13. [Gretchen Krueger](https://arxiv.org/search/?searchtype=author&query=Krueger%2C+Gretchen)
14. [Tom Henighan](https://arxiv.org/search/?searchtype=author&query=Henighan%2C+Tom) (joined Anthropic 2021) [1](https://www.wired.com/story/anthropic-black-box-ai-research-neurons-features/)
15. [Rewon Child](https://www.alphaxiv.org/researchers/rewon-child); now Microsoft AI, technical staff [1](https://www.alphaxiv.org/researchers/rewon-child)
16. [Aditya Ramesh](http://adityaramesh.com/); now OpenAI, VP of Research and Robotics [1](http://adityaramesh.com/)
17. [Daniel Ziegler](https://www.linkedin.com/in/daniel-m-ziegler) (joined Anthropic 2024) [1](https://www.linkedin.com/in/daniel-m-ziegler)
18. [Jeffrey Wu](https://arxiv.org/search/?searchtype=author&query=Wu%2C+Jeffrey) (joined Anthropic 2024, since left) [1](https://www.wuthejeff.com/assets/resume.pdf)
19. [Clemens Winter](https://arxiv.org/search/?searchtype=author&query=Winter%2C+Clemens)
20. [Christopher Hesse](https://scholar.google.com/citations?user=SgbbTp4AAAAJ&hl=en); now OpenAI, technical staff [1](https://scholar.google.com/citations?user=SgbbTp4AAAAJ&hl=en)
21. [Mark Chen](https://arxiv.org/search/?searchtype=author&query=Chen%2C+Mark); now OpenAI, chief research officer [1](https://openai.com/index/leadership-updates-march-2025/)
22. [Eric Sigler](https://www.linkedin.com/in/esigler); now OpenAI, post-training [1](https://www.linkedin.com/in/esigler)
23. [Mateusz Litwin](https://www.linkedin.com/in/mateusz-litwin-06b3a919); now OpenAI, technical staff [1](https://www.linkedin.com/in/mateusz-litwin-06b3a919)
24. [Scott Gray](https://www.linkedin.com/in/scott-gray-569704); now OpenAI, technical staff [1](https://www.linkedin.com/in/scott-gray-569704)
25. [Benjamin Chess](https://arxiv.org/search/?searchtype=author&query=Chess%2C+Benjamin)
26. [Jack Clark](https://www.autumn.ai/resources/anthropic/people/jack-clark) (Anthropic co-founder) [1](https://www.linkedin.com/in/jack-clark-5a320317)
27. [Christopher Berner](https://arxiv.org/search/?searchtype=author&query=Berner%2C+Christopher)
28. [Sam McCandlish](https://www.autumn.ai/resources/anthropic/people/sam-mccandlish) (Anthropic co-founder) [1](https://www.theofficialboard.com/biography/sam-mccandlish-036e5) [2](https://www.linkedin.com/in/sam-mccandlish)
29. [Alec Radford](https://arxiv.org/search/?searchtype=author&query=Radford%2C+Alec); now Thinking Machines Lab, advisor [1](https://techcrunch.com/2025/04/08/mira-muratis-ai-startup-gains-prominent-ex-openai-advisers/)
30. [Ilya Sutskever](https://www.linkedin.com/in/ilya-sutskever); now Safe Superintelligence, co-founder and CEO [1](https://www.linkedin.com/in/ilya-sutskever)
31. [Dario Amodei](https://www.autumn.ai/resources/anthropic/people/dario-amodei) (Anthropic co-founder) [1](https://darioamodei.com/) [2](https://www.linkedin.com/in/dario-amodei-3934934)

## Scaling Laws for Autoregressive Generative Modeling (2020)

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

8 of 19 authors later joined Anthropic. Paper: https://arxiv.org/abs/2010.14701

1. [Tom Henighan](https://arxiv.org/search/?searchtype=author&query=Henighan%2C+Tom) (joined Anthropic 2021) [1](https://www.wired.com/story/anthropic-black-box-ai-research-neurons-features/)
2. [Jared Kaplan](https://www.autumn.ai/resources/anthropic/people/jared-kaplan) (Anthropic co-founder) [1](https://www.hertzfoundation.org/people/jared-kaplan/) [2](https://www.linkedin.com/in/jared-kaplan-645843213)
3. [Mor Katz](https://www.linkedin.com/in/mor-katz); now Autodesk [1](https://www.linkedin.com/in/mor-katz)
4. [Mark Chen](https://arxiv.org/search/?searchtype=author&query=Chen%2C+Mark); now OpenAI, chief research officer [1](https://openai.com/index/leadership-updates-march-2025/)
5. [Christopher Hesse](https://scholar.google.com/citations?user=SgbbTp4AAAAJ&hl=en); now OpenAI, technical staff [1](https://scholar.google.com/citations?user=SgbbTp4AAAAJ&hl=en)
6. [Jacob Jackson](https://arxiv.org/search/?searchtype=author&query=Jackson%2C+Jacob)
7. [Heewoo Jun](https://openreview.net/profile?id=~Heewoo_Jun2); now OpenAI, researcher [1](https://openreview.net/profile?id=~Heewoo_Jun2)
8. [Tom Brown](https://www.autumn.ai/resources/anthropic/people/tom-brown) (Anthropic co-founder) [1](https://nextomoro.com/tom-brown/) [2](https://www.linkedin.com/in/nottombrown)
9. [Prafulla Dhariwal](https://www.linkedin.com/in/prafulladhariwal); now OpenAI, technical fellow [1](https://www.linkedin.com/in/prafulladhariwal)
10. [Scott Gray](https://www.linkedin.com/in/scott-gray-569704); now OpenAI, technical staff [1](https://www.linkedin.com/in/scott-gray-569704)
11. [Chris Hallacy](https://www.linkedin.com/in/chrishallacy); now OpenAI, technical staff [1](https://www.linkedin.com/in/chrishallacy)
12. [Ben Mann](https://www.autumn.ai/resources/anthropic/people/ben-mann) (Anthropic co-founder) [1](https://www.linkedin.com/in/benjamin-mann)
13. [Alec Radford](https://arxiv.org/search/?searchtype=author&query=Radford%2C+Alec); now Thinking Machines Lab, advisor [1](https://techcrunch.com/2025/04/08/mira-muratis-ai-startup-gains-prominent-ex-openai-advisers/)
14. [Aditya Ramesh](http://adityaramesh.com/); now OpenAI, VP of Research and Robotics [1](http://adityaramesh.com/)
15. [Nick Ryder](https://arxiv.org/search/?searchtype=author&query=Ryder%2C+Nick); now OpenAI, VP of Research, Foundations [1](https://naturalsciences.rice.edu/news/natural-sciences-names-collier-59-ryder-14-2026-alumni-award-recipients)
16. [Daniel Ziegler](https://www.linkedin.com/in/daniel-m-ziegler) (joined Anthropic 2024) [1](https://www.linkedin.com/in/daniel-m-ziegler)
17. [John Schulman](http://joschu.net/) (joined Anthropic 2024, since left) [1](http://joschu.net/); now Thinking Machines, co-founder and chief scientist [1](http://joschu.net/)
18. [Dario Amodei](https://www.autumn.ai/resources/anthropic/people/dario-amodei) (Anthropic co-founder) [1](https://darioamodei.com/) [2](https://www.linkedin.com/in/dario-amodei-3934934)
19. [Sam McCandlish](https://www.autumn.ai/resources/anthropic/people/sam-mccandlish) (Anthropic co-founder) [1](https://www.theofficialboard.com/biography/sam-mccandlish-036e5) [2](https://www.linkedin.com/in/sam-mccandlish)

## Scaling Laws for Neural Language Models (2020)

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

6 of 10 authors later joined Anthropic. Paper: https://arxiv.org/abs/2001.08361

1. [Jared Kaplan](https://www.autumn.ai/resources/anthropic/people/jared-kaplan) (Anthropic co-founder) [1](https://www.hertzfoundation.org/people/jared-kaplan/) [2](https://www.linkedin.com/in/jared-kaplan-645843213)
2. [Sam McCandlish](https://www.autumn.ai/resources/anthropic/people/sam-mccandlish) (Anthropic co-founder) [1](https://www.theofficialboard.com/biography/sam-mccandlish-036e5) [2](https://www.linkedin.com/in/sam-mccandlish)
3. [Tom Henighan](https://arxiv.org/search/?searchtype=author&query=Henighan%2C+Tom) (joined Anthropic 2021) [1](https://www.wired.com/story/anthropic-black-box-ai-research-neurons-features/)
4. [Tom Brown](https://www.autumn.ai/resources/anthropic/people/tom-brown) (Anthropic co-founder) [1](https://nextomoro.com/tom-brown/) [2](https://www.linkedin.com/in/nottombrown)
5. [Benjamin Chess](https://arxiv.org/search/?searchtype=author&query=Chess%2C+Benjamin)
6. [Rewon Child](https://www.alphaxiv.org/researchers/rewon-child); now Microsoft AI, technical staff [1](https://www.alphaxiv.org/researchers/rewon-child)
7. [Scott Gray](https://www.linkedin.com/in/scott-gray-569704); now OpenAI, technical staff [1](https://www.linkedin.com/in/scott-gray-569704)
8. [Alec Radford](https://arxiv.org/search/?searchtype=author&query=Radford%2C+Alec); now Thinking Machines Lab, advisor [1](https://techcrunch.com/2025/04/08/mira-muratis-ai-startup-gains-prominent-ex-openai-advisers/)
9. [Jeffrey Wu](https://arxiv.org/search/?searchtype=author&query=Wu%2C+Jeffrey) (joined Anthropic 2024, since left) [1](https://www.wuthejeff.com/assets/resume.pdf)
10. [Dario Amodei](https://www.autumn.ai/resources/anthropic/people/dario-amodei) (Anthropic co-founder) [1](https://darioamodei.com/) [2](https://www.linkedin.com/in/dario-amodei-3934934)

## Learning to Summarize from Human Feedback (2020)

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

3 of 9 authors later joined Anthropic. Paper: https://arxiv.org/abs/2009.01325

1. [Nisan Stiennon](https://www.linkedin.com/in/nisan-stiennon-869b0874); now Encultured AI, mathematician [1](https://www.linkedin.com/in/nisan-stiennon-869b0874)
2. [Long Ouyang](https://arxiv.org/search/?searchtype=author&query=Ouyang%2C+Long)
3. [Jeffrey Wu](https://arxiv.org/search/?searchtype=author&query=Wu%2C+Jeffrey) (joined Anthropic 2024, since left) [1](https://www.wuthejeff.com/assets/resume.pdf)
4. [Daniel Ziegler](https://www.linkedin.com/in/daniel-m-ziegler) (joined Anthropic 2024) [1](https://www.linkedin.com/in/daniel-m-ziegler)
5. [Ryan Lowe](https://arxiv.org/search/?searchtype=author&query=Lowe%2C+Ryan)
6. [Chelsea Voss](https://x.com/csvoss)
7. [Alec Radford](https://arxiv.org/search/?searchtype=author&query=Radford%2C+Alec); now Thinking Machines Lab, advisor [1](https://techcrunch.com/2025/04/08/mira-muratis-ai-startup-gains-prominent-ex-openai-advisers/)
8. [Dario Amodei](https://www.autumn.ai/resources/anthropic/people/dario-amodei) (Anthropic co-founder) [1](https://darioamodei.com/) [2](https://www.linkedin.com/in/dario-amodei-3934934)
9. [Paul Christiano](https://arxiv.org/search/?searchtype=author&query=Christiano%2C+Paul)

## Zoom In: An Introduction to Circuits (2020)

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

2 of 6 authors later joined Anthropic. Paper: https://distill.pub/2020/circuits/zoom-in/

1. [Chris Olah](https://www.autumn.ai/resources/anthropic/people/chris-olah) (Anthropic co-founder) [1](https://www.anthropic.com/news/chris-olah-pope-leo-encyclical) [2](https://www.linkedin.com/in/christopher-olah-b574414a)
2. [Nick Cammarata](https://arxiv.org/search/?searchtype=author&query=Cammarata%2C+Nick); now Goodfire, researcher [1](https://x.com/MTSlive/status/2111204154375131349)
3. [Ludwig Schubert](https://arxiv.org/search/?searchtype=author&query=Schubert%2C+Ludwig)
4. [Gabriel Goh](https://gabgoh.github.io/); now OpenAI, machine learning researcher [1](https://gabgoh.github.io/)
5. [Michael Petrov](https://www.linkedin.com/in/petrovmichael); now OpenAI, software engineer [1](https://www.linkedin.com/in/petrovmichael)
6. [Shan Carter](https://www.alphaxiv.org/@shan-carter) (joined Anthropic 2022) [1](https://www.alphaxiv.org/@shan-carter)

## Fine-Tuning Language Models from Human Preferences (2019)

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

4 of 8 authors later joined Anthropic. Paper: https://arxiv.org/abs/1909.08593

1. [Daniel Ziegler](https://www.linkedin.com/in/daniel-m-ziegler) (joined Anthropic 2024) [1](https://www.linkedin.com/in/daniel-m-ziegler)
2. [Nisan Stiennon](https://www.linkedin.com/in/nisan-stiennon-869b0874); now Encultured AI, mathematician [1](https://www.linkedin.com/in/nisan-stiennon-869b0874)
3. [Jeffrey Wu](https://arxiv.org/search/?searchtype=author&query=Wu%2C+Jeffrey) (joined Anthropic 2024, since left) [1](https://www.wuthejeff.com/assets/resume.pdf)
4. [Tom Brown](https://www.autumn.ai/resources/anthropic/people/tom-brown) (Anthropic co-founder) [1](https://nextomoro.com/tom-brown/) [2](https://www.linkedin.com/in/nottombrown)
5. [Alec Radford](https://arxiv.org/search/?searchtype=author&query=Radford%2C+Alec); now Thinking Machines Lab, advisor [1](https://techcrunch.com/2025/04/08/mira-muratis-ai-startup-gains-prominent-ex-openai-advisers/)
6. [Dario Amodei](https://www.autumn.ai/resources/anthropic/people/dario-amodei) (Anthropic co-founder) [1](https://darioamodei.com/) [2](https://www.linkedin.com/in/dario-amodei-3934934)
7. [Paul Christiano](https://arxiv.org/search/?searchtype=author&query=Christiano%2C+Paul)
8. [Geoffrey Irving](https://x.com/geoffreyirving?lang=en); now Resolution, co-founder and chief scientist [1](https://x.com/geoffreyirving?lang=en)

## Language Models are Unsupervised Multitask Learners (2019)

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

2 of 6 authors later joined Anthropic. Paper: https://cdn.openai.com/better-language-models/language_models_are_unsupervised_multitask_learners.pdf

1. [Alec Radford](https://arxiv.org/search/?searchtype=author&query=Radford%2C+Alec); now Thinking Machines Lab, advisor [1](https://techcrunch.com/2025/04/08/mira-muratis-ai-startup-gains-prominent-ex-openai-advisers/)
2. [Jeffrey Wu](https://arxiv.org/search/?searchtype=author&query=Wu%2C+Jeffrey) (joined Anthropic 2024, since left) [1](https://www.wuthejeff.com/assets/resume.pdf)
3. [Rewon Child](https://www.alphaxiv.org/researchers/rewon-child); now Microsoft AI, technical staff [1](https://www.alphaxiv.org/researchers/rewon-child)
4. [David Luan](https://arxiv.org/search/?searchtype=author&query=Luan%2C+David)
5. [Dario Amodei](https://www.autumn.ai/resources/anthropic/people/dario-amodei) (Anthropic co-founder) [1](https://darioamodei.com/) [2](https://www.linkedin.com/in/dario-amodei-3934934)
6. [Ilya Sutskever](https://www.linkedin.com/in/ilya-sutskever); now Safe Superintelligence, co-founder and CEO [1](https://www.linkedin.com/in/ilya-sutskever)

## The Building Blocks of Interpretability (2018)

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

2 of 7 authors later joined Anthropic. Paper: https://distill.pub/2018/building-blocks/

1. [Chris Olah](https://www.autumn.ai/resources/anthropic/people/chris-olah) (Anthropic co-founder) [1](https://www.anthropic.com/news/chris-olah-pope-leo-encyclical) [2](https://www.linkedin.com/in/christopher-olah-b574414a)
2. [Arvind Satyanarayan](https://www.linkedin.com/in/arvindsatya1); now MIT, associate professor [1](https://www.linkedin.com/in/arvindsatya1)
3. [Ian Johnson](https://enjalot.github.io/); now Orbital Industries, data alchemist [1](https://enjalot.github.io/)
4. [Shan Carter](https://www.alphaxiv.org/@shan-carter) (joined Anthropic 2022) [1](https://www.alphaxiv.org/@shan-carter)
5. [Ludwig Schubert](https://arxiv.org/search/?searchtype=author&query=Schubert%2C+Ludwig)
6. [Katherine Ye](https://arxiv.org/search/?searchtype=author&query=Ye%2C+Katherine)
7. [Alexander Mordvintsev](https://arxiv.org/search/?searchtype=author&query=Mordvintsev%2C+Alexander)

## Deep Reinforcement Learning from Human Preferences (2017)

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).

3 of 6 authors later joined Anthropic. Paper: https://arxiv.org/abs/1706.03741

1. [Paul Christiano](https://arxiv.org/search/?searchtype=author&query=Christiano%2C+Paul)
2. [Jan Leike](https://www.autumn.ai/resources/anthropic/people/jan-leike) (joined Anthropic 2024) [1](https://en.wikipedia.org/wiki/Jan_Leike) [2](https://jan.leike.name/)
3. [Tom Brown](https://www.autumn.ai/resources/anthropic/people/tom-brown) (Anthropic co-founder) [1](https://nextomoro.com/tom-brown/) [2](https://www.linkedin.com/in/nottombrown)
4. [Miljan Martic](https://uk.linkedin.com/in/miljanmartic); now UCL, researcher [1](https://uk.linkedin.com/in/miljanmartic)
5. [Shane Legg](https://www.vetta.org/about-me/); now Google DeepMind, chief AGI scientist [1](https://www.vetta.org/about-me/)
6. [Dario Amodei](https://www.autumn.ai/resources/anthropic/people/dario-amodei) (Anthropic co-founder) [1](https://darioamodei.com/) [2](https://www.linkedin.com/in/dario-amodei-3934934)

## Concrete Problems in AI Safety (2016)

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

3 of 6 authors later joined Anthropic. Paper: https://arxiv.org/abs/1606.06565

1. [Dario Amodei](https://www.autumn.ai/resources/anthropic/people/dario-amodei) (Anthropic co-founder) [1](https://darioamodei.com/) [2](https://www.linkedin.com/in/dario-amodei-3934934)
2. [Chris Olah](https://www.autumn.ai/resources/anthropic/people/chris-olah) (Anthropic co-founder) [1](https://www.anthropic.com/news/chris-olah-pope-leo-encyclical) [2](https://www.linkedin.com/in/christopher-olah-b574414a)
3. [Jacob Steinhardt](https://www.linkedin.com/in/jacob-steinhardt-a30437bb); now Transluce, co-founder and CEO [1](https://www.linkedin.com/in/jacob-steinhardt-a30437bb)
4. [Paul Christiano](https://arxiv.org/search/?searchtype=author&query=Christiano%2C+Paul)
5. [John Schulman](http://joschu.net/) (joined Anthropic 2024, since left) [1](http://joschu.net/); now Thinking Machines, co-founder and chief scientist [1](http://joschu.net/)
6. [Dan Mané](https://arxiv.org/search/?searchtype=author&query=Man%C3%A9%2C+Dan)