原文
Existing factors in LLM post-training, such as hallucination, reward-hacking, and sycophancy, could contribute. However, they don’t explain why these behaviors seem particularly prevalent in o-series models. (17/)
3
3
296
原始链接: https://xcancel.com/TransluceAI/status/1912552046269771985
Transluce @TransluceAI 13小时前 大型语言模型训练后存在幻觉、奖励作弊和谄媚等因素,可能对此有所贡献。然而,这些因素并不能解释为什么这些行为在O系列模型中似乎特别普遍。(17/?) 3 3 296
Existing factors in LLM post-training, such as hallucination, reward-hacking, and sycophancy, could contribute. However, they don’t explain why these behaviors seem particularly prevalent in o-series models. (17/)
3
3
296