Main limiting factor? Hard to know. But throw 100x more compute to an LLM (through an increase of both parameters and tokens) and it will undoubtedly beat curr

Main limiting factor? Hard to know. But throw 100x more compute to an LLM (through an increase of both parameters and tokens) and it will undoubtedly beat current models.