Hacker News new | ask | show | jobs
by orbital-decay 1 hour ago
>But if I ask an LLM 'who are the best commercial roofers in Houston?', there isn't one right answer. When an LLM answers that roofing question, it typically cites 3 to 5 businesses. The other 40 legitimate roofing companies in the area are left out.

I don't see how mentioning the other 40 should follow from that fact. You get exactly what you're asking for - the best, according to agent's judgement. Judgement and processing of simple search results is exactly what you're using the intelligent agent for. Ask for a list of all commercial roofers in Houston and then you can expect a list.

It can be argued that models are clustering their replies around a few options when asked to choose from a list of equally valid ones (due to mode collapse, distribution biases, primacy/recency biases etc), but these options are not equally valid for the agent, it looks at their sites or possibly in some other places like review sites to rank them. Its ranking criteria might be not good at all, but that's another question.