Although this has been discussed in the Hutter Prize FAQ for many years, when the OpenAI Chief Scientist discusses why lossless compression is the most principled loss function, it may harness the LLM stampede and put some of those billions flying around to good use in answering hard questions about "bias" not only in ML, but in data-driven sciences.
