@ESYudkowsky @mattyglesias gpt-4o is a significant advance wrt intelligence per flop; we need a standard metric for this and for OpenAI & other closed model providers to publish info wrt flops-per-output-token so we can compare them & have energy-adjusted leaderboards