| People get so distracted trying to use certain significant words for what LLM’s do, even when the usage is strained and makes it harder to see how they actually work and what they excel at. A better word for what they do here might be something like “preambulating” — it develops a focus to its later output by grounding more and more tokens into its active context, because they each narrow what else fits. That winnowing effect helps it produce a coherent and rich answer, and when you undermine its opportunity to use that technique, the answers become less coherent and more random. This is not reasoning as that word is traditionally used and doesn’t need to be called that. Yet it’s still a fascinating emergent phenomenon with incredible engineering opportunity. When you call it by something less culturally ambitious and more technically precise, it helps you stay focused on how to use it well and less distracted by some personal desire to prove this is the exact historical moment you want it to be. We need to develop a better vocabulary around these things if we want to stop having the dumb Nascent AGI vs Fancy Autocomplete flamewar. Edit: And I’ll even throw a bone to the Nascent AGI people and say that this kind of preambulating is absolutely something that people do too and easy to characterize as some form of intelligence. But it’s not reasoning, which has specific strong connotations of formality and logic, which don’t hold well with these particular tools. |