3 ms·
No, they just need to be trained to have adversarial self review "thinking" processes. You ask an LLM "What's wrong with your answer?" and you get pretty good
by colechristensen 6mo ago
No, they just need to be trained to have adversarial self review "thinking" processes.
You ask an LLM "What's wrong with your answer?" and you get pretty good results.
- binary0010 6mo agoOr you get the original output result was perfect and the adversarial "rethinking" switches to an incorrect result.
- byzantinegene 6mo agothis seems to happen far more than i would like