|
|
|
|
|
by simianwords
316 days ago
|
|
Your internal verifier model in your head is actually good enough and not random. It knows how the world works and subconsciously applies a lot of sniff tests it has learned over the years. Sure a lot of answers from llms may be inaccurate - but you mostly identify them as such because your ability to verify (using various heuristics) is good too. Do you learn from asking people advice? Do you learn from reading comments on Reddit? You still do without trusting them fully because you have sniff tests. |
|
LLMs produce way too much noise and way too inconsistent quality for a sniff test to be terribly valuable in my opinion