3 ms·
> that conforms precisely to what's been asked This. People talks about prompt engineering, but then it fails on really simple details, like "on lowercase",
by txutxu 3y ago
> that conforms precisely to what's been asked
This.
People talks about prompt engineering, but then it fails on really simple details, like "on lowercase", "composed by max two words", etc... and when you point at the failure, apologizes, and composes something else that forgets the other 95% of the original prompt.
Or worse, apologizes and makes again the very same mistake.
- skywhopper 3y agoThis sucks, but it's unlikely to be fixable, given that LLMs don't actually have any comprehension or reasoning capability. Get too far into fine-tuning responses and you're back to "classic" AI problems.