OpenAI will reportedly ship at least one model this year but its flagship still needs alignment work.

OpenAI is working to finish its next flagship model, which uses a looped transformer technique like Astra to improve math and coding performance. The looping approach reportedly makes safety monitoring harder.

Why it matters: The technique improves math and coding ability but complicates the safety monitoring needed before the flagship model can ship.

Written by AI from The Decoder's published text. Read the original for full details.