ML models to teach the AI bot to overthink?
As i see, "attention is all you need" but not quite. As a human, we often overthink, too. So my expectation is current AI bots should think humanly by overthink or overabstract things, too. How do you think and what do you think of this ?
Overthinking produces more junk text which overall clutters the model's context window and reduces the effectiveness of attention in the first place. It's autocomplete - if you precede it with a bunch of junk data, you're going to get junk suggestions back.
LLMs can't and don't use logic/reason or "think," so the analogy to human overthinking doesn't make sense.
Overthinking is often driven by emotion, which machines can't feel either.