Hacker News Reader

So it sounds like the proposed use-case is to use LLMs for generating feedback on your own work.

But if we accept that LLMs generally (in other use-cases) produce output that looks deceptively similar to what you ask for (i.e. it seems to work) but is actually worthless junk if carefully inspected (i.e. it doesn't actually work), why would you think they are able to generate accurate feedback?

View on HN

The problem with deceptively correct information is that it leads us to take mental shortcuts that lead us to oversee details. If you do the work first, you've already avoided this problem. Instead, if the feedback of the llm doesn't match your mental model, it leads you to think twice about it, which is a good thing (the risk of overthinking is offset by the speed at which llms give you the feedback).

Reply on news.ycombinator.com