Nicholas Joseph
Source-listed: Head of Pretraining · Anthropic
Record snapshot: Sep 3, 2026. Coverage and affiliations may be incomplete or historical.
Selected work
5- Constitutional AI: Harmlessness from AI FeedbackSaved credit: Contributor · 2022Explore people connected to this work →
- Training a Helpful and Harmless Assistant with Reinforcement Learning from Human FeedbackSaved credit: Contributor · 2022Explore people connected to this work →
- Predictability and Surprise in Large Generative ModelsSaved credit: Contributor · 2022Explore people connected to this work →
- Discovering Language Model Behaviors with Model-Written EvaluationsSaved credit: Contributor · 2022Explore people connected to this work →
- Question Decomposition Improves the Faithfulness of Model-Generated ReasoningSaved credit: Contributor · 2023Explore people connected to this work →