OpenAI launches a science blog
OpenAI has started a blog where scientists can report what they discover with AI and how they do it, as the company points to recent claims about results in mathematics and biology.
OpenAI announced a new science blog for scientists and academic researchers to share what they are finding with AI and how they are doing the work. The company called faster scientific discovery "one of the most impactful ways" to pursue its AGI mission.
The blog will cover three kinds of material:
- Research results: new findings and progress updates from OpenAI, its collaborators and the wider research community.
- Methods: how scientists use OpenAI models for scientific computing, data analysis, reading the literature and planning experiments. OpenAI says this will include how to check the work and where a method falls short.
- Researcher perspectives: firsthand accounts from people using AI in their work, and how it is changing their fields.
OpenAI used the launch to list recent claims. It said an internal system produced a proof in September "addressing" the Navier–Stokes existence and smoothness problem. It also pointed to a result in October on what it called the quasi-Riemann hypothesis, which sharpens estimates of how many primes fall below a given number. The post does not say whether outside mathematicians have verified either result.
In biology, OpenAI said that in work with PerturbAI and Scripps Research, thousands of agents running on GPT-Rosalind analyzed existing data from millions of mouse brain cells in 32 minutes. The agents proposed possible biological mechanisms for scientists to test in the lab. OpenAI also cited César de la Fuente's team, which uses ChatGPT and Codex to find potential antibiotic molecules that the team then makes and tests.
OpenAI acknowledged open questions about how AI-generated proofs are checked, how researchers choose which hypotheses to test, and how these tools affect training and credit.