Mirendil, a startup founded by veterans of Anthropic and other top AI labs, raised $200M in seed funding at a $1B valuation from Andreessen Horowitz, Kleiner Perkins, and Nvidia, one of the larger seed valuations for a new AI company in recent years. The company is building AI that accelerates AI research itself, aiming to let open-source developers and scientists develop their own specialized in-house models in fields like medicine and materials, which co-founder Behnam Neyshabur described as “AI for AI for science.” The bet directly targets a gap left by frontier labs, which use AI to speed their own research but bar outsiders from using their tools to compete. Mirendil's founders see recursive self-improvement as the shortest path to accelerating science, a position that runs against AI-safety researchers who view models rewriting their own code as a potential danger.
