Jan Leike
Leads the Alignment Science team at Anthropic; formerly co-led Superalignment at OpenAI
Jan Leike is a machine learning and alignment researcher who leads the Alignment Science team at Anthropic, where the team works on aligning an automated alignment researcher, scalable oversight, weak-to-strong generalization and robustness to jailbreaks. Leike previously co-led the Superalignment team at OpenAI and was involved in developing InstructGPT, ChatGPT and the alignment of GPT-4; before that, Leike was an alignment researcher at DeepMind who prototyped reinforcement learning from human feedback. Leike is second author of "Deep reinforcement learning from human preferences" (2017) and first author of "Scalable agent alignment via reward modeling" (2018).123
Work
Deep reinforcement learning from human preferences (external site: arxiv.org)Published research
Second-listed author of the 2017 paper that trains agents from human preferences between pairs of trajectory segments.2
Scalable agent alignment via reward modeling (external site: arxiv.org)Published research
First author of the 2018 research-direction paper on aligning agents through reward modeling.3
InstructGPT and weak-to-strong generalization (external site: arxiv.org)Published research
Co-author of "Training language models to follow instructions with human feedback" (2022) and of "Weak-to-Strong Generalization: Eliciting Strong Capabilities With Weak Supervision" (2023).45
Other documented roles
Roles the sources document but that are not confirmed as current.
This profile covers public professional work only, from the sources below. It is not an endorsement or a ranking. To correct or remove it, use the correction form.