What is Discovery Loop, Chinese teen AI prodigy Nathan Chen’s rumoured next destination?

Chinese teen AI prodigy Nathan Chen Guangyu, who recently left his role at Moonshot AI, appears to have secured a new position at Discovery Loop, a rapidly growing AI startup. Discovery Loop was established by a group of former Google senior researchers, including the former chief scientist Jeff Dean. The startup has already drawn significant attention due to its impressive team and mission to automate scientific research using AI.

Since its inception just two months ago, Discovery Loop has already attracted several top researchers from major US tech companies, including other Google engineers. Despite being only 17 years old, Chen is expected to be one of the youngest members of the team. While Chen did not immediately respond to requests for comment, Discovery Loop declined to comment on personnel matters.

Chen's decision to leave China for the US last week had sparked speculation about his next destination. During his time at Moonshot, which lasted less than a year, Chen contributed to significant technical advancements that made Moonshot's flagship Kimi models competitive with US AI leaders such as OpenAI's GPT and Anthropic's Claude.

Moonshot's K3 model has gained global recognition due to its powerful capabilities and open-weight approach. Following Chen's announcement, Moonshot's senior researcher Du Yulun shared that Chen and he had long discussions about his decision to leave, but ultimately, it was in Chen's best interest to pursue new curiosities at his young age.

Discovery Loop is currently seeking a valuation of approximately US$50 billion in its latest funding round, with initial investments led by Radical Ventures, Khosla Ventures, and Alphabet, Google's parent company. Dean recently shared his startup's work would aim to achieve similar results as those seeking recursive self-improvement (RSI), which involves automating AI development through AI itself.

Dean also explained that AI could eventually run thousands of scientific experiments in parallel, significantly reducing the time required for scientific progress from years to mere hours or even minutes by autonomously identifying the most promising experiments based on the results.