These AI Experts Want to Do High-Stakes Research Out in the Open

A nonprofit organization founded by two AI researchers wants to conduct high-stakes artificial intelligence research in the open by publishing experimental details so outside scientists can study and replicate the work. The founders argue that major AI companies keep their research secret to manage risks, but this approach prevents the broader scientific community from scrutinizing ideas and contributing solutions. The nonprofit plans to focus on potentially problematic areas including recursive self-improvement, agents, post-training methods, and reinforcement learning. This reflects an ongoing debate in the industry about whether powerful AI models are safer when kept behind closed doors in trusted labs or when developed transparently with wider scientific input.