Andrej Karpathy, a prominent AI researcher and co-founder of OpenAI, has recently joined Anthropic. His new role focuses on pre-training, a critical phase in developing AI systems.
In a post on X, Karpathy expressed his enthusiasm for the future of large language models (LLMs), stating, "I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D." He also mentioned his ongoing passion for education and plans to return to that work in due time.
Karpathy began his position at Anthropic this week, working under team lead Nick Joseph. Pre-training is essential for large-scale training runs that equip AI models like Claude with foundational knowledge and capabilities. This phase is known for being both compute-intensive and costly.
According to an Anthropic spokesperson, Karpathy will establish a team aimed at leveraging Claude to enhance pre-training research. His expertise uniquely positions him to bridge theoretical aspects of LLMs with practical training applications.
Karpathy's hiring indicates Anthropic's commitment to AI-assisted research as a competitive strategy against industry leaders like OpenAI and Google. Previously, he led deep learning and computer vision efforts at OpenAI before moving to Tesla, where he managed the Full Self-Driving and Autopilot programs until 2022.
After a brief return to OpenAI, Karpathy launched Eureka Labs, a startup focused on integrating AI into education. His recent silence regarding Eureka Labs raises questions about its future direction.
In addition to his new role, Karpathy has also been involved in educational initiatives, including an online course on neural networks and a YouTube channel featuring lectures on LLMs and AI.
In a related development, Anthropic has also welcomed Chris Rohlf to its frontier red team, which is tasked with stress-testing advanced AI models. Rohlf brings over 20 years of cybersecurity experience, having previously worked at Yahoo and Meta.
Rohlf remarked on the potential for AI to significantly enhance cybersecurity, stating, "We have a real opportunity in front of us to dramatically improve cyber security with AI."
