OpenAI says automated research could improve alignment, but "fully autonomous RSI is not happening today" and shouldn't be pursued unless it can be done safely (OpenAI)

OpenAI announced that automated research can enhance alignment but that fully autonomous research-synthesis initiatives are not currently feasible and should not be pursued until they can be performed safely. The company is collaborating with an independent math advisory group and has begun training a new internal model that, among other tasks, aims to solve the Navier-Stokes Millennium Prize problem.