4 ms·
Do you have any references? I know of the emergent deception problem that seems to be created through feedback. https://bounded-regret.ghost.io/emergent-decept
by 13years 3y ago
Do you have any references? I know of the emergent deception problem that seems to be created through feedback.
https://bounded-regret.ghost.io/emergent-deception-optimization https://bounded-regret.ghost.io/emergent-deception-optimizat...
- famouswaffles 3y agoBase GPT-4 was highly calibrated. read open ai's technical paper. also, this paper on gpt-4 performance of medical challenge problems confirmed the high calibration for medicine https://arxiv.org/abs/2303.13375 https://arxiv.org/abs/2303.13375
- 13years 3y agoThanks, but I didn't find any details about performance of pre reinforcement training and after. Looking to understand more about the assertion that hallucinations are introduced by the reinforcement training.
- famouswaffles 3y agohttps://arxiv.org/abs/2303.08774 https://arxiv.org/abs/2303.08774 The technical report has before and after comparisons. It's a bit worse on some tests. and they pretty explicitly mention the issue of calibration (how well confidence on a problem results in the ability or accuracy solving that problem). Hallucinations are a different matter.