Particle.news
Download on the App Store
29 ARTICLES
·
4w ago

Andrej Karpathy Joins Anthropic to Lead Claude Pre‑Training Research

The hire signals Anthropic will lean on its Claude model to speed costly pre-training work, bolstering the company’s research throughput and security testing capacity.

Andrej Karpathy, when he was a director of AI at Tesla, speaking at a conference on Thurs. May 10, 2018.
Image
Illustration: Sarah Grillo/Axios. Stock: Getty Images

Overview

  • Andrej Karpathy, a founding member of OpenAI and former Tesla AI director, joined Anthropic on Tuesday, May 19 to start work on its pre-training team.
  • Anthropic says Karpathy will build a new group that uses Claude to automate and accelerate pre-training, the expensive phase that sets a foundation model’s capabilities.
  • Karpathy will report to pre-training lead Nick Joseph and focus on AI-assisted research methods that aim to shorten experiment cycles without only adding raw compute.
  • Anthropic also hired cybersecurity veteran Chris Rohlf for its frontier red team to expand adversarial testing and harden models against misuse.
  • The moves strengthen Anthropic’s push to scale Claude through talent, tooling and large compute partnerships and sharpen competition for top researchers across leading AI labs.