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

OpenAI is developing a new flagship model that utilizes a looped transformer technique to enhance capabilities in mathematics and programming. While the company intends to release at least one new model this year, the integration of this technique has complicated safety monitoring efforts. This development marks a shift in architectural strategy, though it forces a trade-off between increased performance and the time required for final alignment testing.