Renowned artificial intelligence researcher Andrej Karpathy, co-founder of OpenAI and former leader of Tesla's AI division, has just confirmed his joining Anthropic (AI company behind chatbot Claude).
This move is seen as a remarkable step forward for Anthropic in the increasingly fierce AI competition with OpenAI and Google.
On social network X, Karpathy said that he has officially joined Anthropic and will return to research and development of major language models (LLM).
According to Anthropic, Karpathy started working this week at the pre-training group led by Nick Joseph.
This is the department responsible for large-scale training sessions to help Claude build foundational knowledge and core competencies.
Pre-training is considered one of the most expensive steps and requires the largest computing resource in the process of developing advanced AI models.
Anthropic also said that Karpathy will build a team specializing in using Claude to support and accelerate pre-training research activities.
Technology experts assess Karpathy as one of the few experts capable of connecting the theory of large-scale language modeling and the reality of AI training on an extremely large scale.
Anthropic's recruitment shows that the company is betting on research supported by AI, instead of just relying on expanding computing capabilities.
Before joining Anthropic, Karpathy worked at OpenAI for many years with a focus on in-depth learning and computer vision.
In 2017, he left OpenAI to join Tesla, where he led the Autopilot and Full Self-Driving (FSD) programs, two core projects related to self-driving cars of this electric vehicle company.
After leaving Tesla in 2022, Karpathy returned to OpenAI for about a year before continuing to leave in 2024 to establish Eureka Labs (startup applying AI assistants in education).
However, since its launch, Karpathy has not shared much new information about Eureka Labs. It is not yet clear whether he will continue to participate in managing this startup.
In addition to AI research, Karpathy is also widely known in the technology community thanks to in-depth courses and lectures on neural networks and large language models. Karpathy also owns a YouTube channel specializing in sharing knowledge about AI and LLM.
Along with recruiting Karpathy, Anthropic also added cybersecurity expert Chris Rohlf to the AI durability testing team (red team). This department is responsible for assessing the resilience of advanced AI models to dangerous threats.
Rohlf has more than 20 years of experience in the field of cybersecurity. He used to work at Yahoo's famous security group "The Paranoids" and had 6 years at Meta before joining Anthropic. In addition, Rohlf also participated in research at the Center for Emerging Security and Technology at Georgetown University.
According to Rohlf, AI could open up great opportunities to improve global cybersecurity and Anthropic is one of the most suitable places to pursue this goal.
