Well, seems like ECI [0] and the AA index is diverging quite a bit. Benchmarking LLM is tough and I think we are seeing the limitations of current benchmarks and applicability to real tasks.

[0]: https://x.com/EpochAIResearch/status/2095602754282783108