OpenAI co-founder Andrej Karpathy joins Anthropic’s pre-training team

Andrej Karpathy, the AI researcher who co-founded and formerly worked astatine OpenAI and antecedently led AI astatine Tesla, has joined Anthropic.

“I’ve joined Anthropic,” Karpathy posted connected X Tuesday. “I deliberation the adjacent fewer years astatine the frontier of LLMs volition beryllium particularly formative. I americium precise excited to articulation the squad present and get backmost to R&D.”

Karpathy started this week astatine Anthropic, wherever helium is moving connected pre-training nether squad pb Nick Joseph. Pre-training is liable for the large-scale grooming runs that springiness Claude its halfway cognition and capabilities, according to the company. It’s besides 1 of the astir expensive, compute-intensive phases of gathering a frontier model.

An Anthropic spokesperson told TechCrunch that Karpathy volition commencement a squad focused connected utilizing Claude to accelerate pre-training research.

Karpathy is 1 of the fewer researchers who tin span the spread betwixt LLM mentation and large-scale grooming practice. Tapping him to physique specified a squad is simply a wide motion from Anthropic that it believes AI-assisted research, alternatively than axenic compute, is however it stays competitory with OpenAI and Google.

While astatine OpenAI, Karpathy focused connected heavy learning and machine imaginativeness until helium departed successful 2017 to articulation Tesla. He led Tesla’s Full Self-Driving (FSD) and Autopilot programs earlier leaving successful 2022.
He past went backmost to OpenAI for 1 twelvemonth earlier leaving again successful 2024 to start Eureka Labs, a startup dedicated to applying AI assistants to education.

Karpathy hasn’t shared galore updates connected Eureka Labs since its launch, and it’s not wide if the renowned researcher volition proceed with the startup. He has besides taught an online people called Neural Networks: Zero to Hero that helps students larn to physique neural networks from scratch successful code, and has a YouTube channel where helium semi-regularly posts lectures connected LLMs and AI. 

“I stay profoundly passionate astir acquisition and program to resume my enactment connected it successful time,” Karpathy said.

TechCrunch has reached retired to Karpathy for comment.

Separately, Anthropic has besides brought connected Chris Rohlf to its frontier reddish team, which stress-tests precocious AI models against terrible threats. Rohlf is simply a seasoned of the cybersecurity manufacture with much than 20 years of experience. He antecedently worked astatine Yahoo’s well-respected cybersecurity squad known arsenic “The Paranoids,” and much precocious astatine Meta, wherever helium worked for six years earlier joining Anthropic. Rohlf was besides a chap astatine Georgetown’s Center for Security and Emerging Technology, wherever helium worked connected the CyberAI project.

“We person a existent accidental successful beforehand of america to dramatically amended cyber information with AI,” Rohlf said successful a post connected X. “I can’t deliberation of a amended institution oregon squad to articulation astatine this captious infinitesimal successful time.”