One of the world’s most influential AI researchers joined the company that makes Claude this week to lead a team that uses AI to speed up the training of models.
One of the world’s most influential AI researchers joined the company that makes Claude this week to lead a team that uses AI to speed up the training of models.
Andrey Karpaty announced his move to Anthropic on Tuesday, May 19, in a post on X. According to TechCrunch, he has already started working on the pre-training team, led by Nick Joseph. Pre-training is the most expensive and resource-intensive phase of building a large language model: this is where Claude gets his basic knowledge and capabilities.
«I think the next few years at the edge of LLM will be particularly defining. I’m very excited to join the team and return to R&D,» Karpaty wrote.
According to an Anthropic representative, Karpaty will form a separate team focused on using Claude to accelerate pre-training research itself. That is, AI will help build better AI — and Karpaty is one of the few in the world who is able to implement this in practice.
Andrey Karpaty was born on October 23, 1986 in Bratislava. He received his bachelor’s degree from the University of Toronto, his master’s degree from the University of British Columbia, and his PhD from Stanford, where he studied under Fei-Fei Li, one of the most influential figures in the field of computer vision.
His doctoral dissertation was about the connection between images and natural language: Karpaty trained a neural network not just to recognize that there was a cat in a photo, but to describe it in detail — «an orange spotted cat on a skateboard with red wheels on a wooden floor.» At Stanford, he also designed and taught CS231n — the first deep learning course with a focus on computer vision, which grew from 150 to 750 students in two years.
In 2015, he became a co-founder and first researcher at OpenAI, before the company became widely known. In 2017, he moved to Tesla, where he spent five years as Senior Director of AI, building the neural network architecture for Tesla Autopilot from scratch, including leading the transition from radar to a camera-only recognition system. Under his leadership, the team fully deployed its own infrastructure: data labeling, neural network training, and deployment on a custom Tesla chip.
After leaving Tesla in 2022, he returned to OpenAI for a year, where he built a new team on midtraining and synthetic data generation. In 2024, he left the company again, this time to found Eureka Labs, an educational AI startup. He also runs a YouTube channel with lectures on neural networks and LLMs that have garnered millions of views, and a free online course called Neural Networks: Zero to Hero.
«I remain deeply passionate about education and plan to return to this work over time,» Karpati noted, commenting on his move to Anthropic.
In parallel, Anthropic announced another addition to its frontier red team: Chris Rolfe, a cybersecurity veteran with over 20 years of experience, has joined the frontier red team. He worked at Yahoo (where the infamous internal security team «The Paranoids» existed), six years at Meta, and was a fellow at Georgetown’s Center for Security and Emerging Technology. The frontier red team at Anthropic is responsible for stress-testing the most powerful models for serious threats.
