LLM output is quickly rendering google search unusable, so it's kind of creating its own speedup multiplier.