EvaLW

·9 bulan yang lalu·discuss

The sensors section of the newsletter is interesting but the model won't see reality, it will see the closest, most plausible pattern it was trained on. It's the physical sciences equivalent of an LLM hallucination.