3 ms·
Of course it's inevitable. Things can be facts or deductions of facts (or both). If I ask an LLM the date of birth of Napoleon and it doesn't have it in its da
by Almondsetat 3y ago
Of course it's inevitable.
Things can be facts or deductions of facts (or both). If I ask an LLM the date of birth of Napoleon and it doesn't have it in its dataset there are only 2 options: either it has other facts from which Napoleon's birthday can be deduced or it doesn't. If it does then by improving the LLM we will be able to make more and more deductions, it if doesn't then it can only hallucinate. Since there will always be a set of facts that the LLM is not aware of and which cannot be deduced we will always have hallucinations
- jtc331 3y agoIs this a way of saying that large language models don’t have the concept of “I don’t know”?
- andybak 3y agoYour "Of course" is a red flag. Unless you have a very precise definition of "LLM" then there's no "of course". It's possible to imagine a hypothetical software system capable of returning "I don't know" or "I'm not sure". I haven't RTFA but I presume they are arguing within specific constraints. The important point is - could an augmented LLM or an "LLM plus something else" solve this. I don't have an answer to that but I don't think it's an "of course" type answer.
- samatman 3y agoI don't see why that's inevitable at all. I immediately recognize that I don't happen to know Napoleon's birthday, and act on that basis.