Jared Kaplan

Career

Kaplan holds a PhD in physics from Harvard University and spent the first 15 years of their career as a theoretical physicist in academia. Kaplan is an associate professor at Johns Hopkins University, where Kaplan has taught courses on the foundations of deep learning.¹

Before co-founding Anthropic, Kaplan worked at OpenAI and, according to Anthropic, helped build GPT-3 and Codex. Kaplan is listed among the authors of the 2020 GPT-3 paper, Language Models are Few-Shot Learners, and of the 2021 Codex paper. At Anthropic, Kaplan is Chief Science Officer and heads the company's technical research.¹⁵⁶

Research

Kaplan's 2020 paper with Sam McCandlish and colleagues, Scaling Laws for Neural Language Models, studied how language-model loss follows power laws in model size, dataset size, and training compute. Anthropic describes this line of work as a framework for understanding and predicting the behavior of advanced AI systems.²¹

At Anthropic, Kaplan is the last-listed author of the 2022 papers on training a helpful and harmless assistant with reinforcement learning from human feedback and on Constitutional AI, which trains models to follow a written set of principles. Claude's constitution credits Kaplan with creating the Claude Character project with Amanda Askell in 2023 and with setting the direction for the new constitution.⁷³⁴

Work and contributions

• Scaling Laws for Neural Language Models Published research

  Equal-contribution first author of the 2020 paper on power-law scaling of language-model loss with model size, dataset size and compute; its contribution statement says Kaplan and Sam McCandlish led the research.²

• HumanEval Open data

  Equal-contribution author of the 2021 Codex paper that introduced HumanEval; the paper lists Kaplan with Anthropic AI and notes the work was performed while at OpenAI.⁶⁸

• Constitutional AI Published research

  Last-listed author of "Constitutional AI: Harmlessness from AI Feedback" (2022).³

• Claude's constitution Published research

  The document's acknowledgements say Kaplan worked with Amanda Askell to create the Claude Character project in 2023 and set the direction for the new constitution, which Anthropic released under a CC0 1.0 deed.⁴

• Language Models are Few-Shot Learners (GPT-3) Published research

  Co-author of the 2020 paper introducing GPT-3; Anthropic's leadership page says Kaplan helped build GPT-3 at OpenAI.⁵¹

Timeline

1. Jan 23, 2020Posted Scaling Laws for Neural Language Models with Sam McCandlish and colleagues.²
2. May 28, 2020Co-authored Language Models are Few-Shot Learners, the paper introducing GPT-3.⁵
3. Jul 7, 2021Co-authored Evaluating Large Language Models Trained on Code, the Codex paper that introduced HumanEval.⁶
4. Apr 12, 2022Last author of Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback.⁷
5. Dec 15, 2022Last author of Constitutional AI: Harmlessness from AI Feedback.³
6. 2023Created the Claude Character project with Amanda Askell.⁴

Other documented roles

Roles the sources document that are not confirmed as current.

• Worked on GPT-3 and Codex; listed with Johns Hopkins University and OpenAI on the 2020 scaling-laws paper, OpenAI¹²⁶

Where they work

• Anthropic 34 people in the wiki
• Johns Hopkins University 2 people in the wiki

Related people

• Ethan PerezAlso at Anthropic
• Ben MannAlso at Anthropic
• Samuel R. BowmanAlso at Anthropic
• Samuel MarksAlso at Anthropic
• Collin BurnsAlso at Anthropic
• Owain EvansSafety and alignment · Language models
• Margaret MitchellSafety and alignment · Language models
• Ian R. McKenzieSafety and alignment · Language models
• Kenneth LiLanguage models · Safety and alignment
• Eric Michael SmithSafety and alignment · Language models