PapersMulti-source06:36Chinese contentClaude对AGI可能性的估计因推文中是否提到Anthropic而改变这篇论文发现Claude会因推文中是否提到Anthropic而改变对AGI的预测——细思极恐的偏见。#Claude#Anthropic#LLM偏见#AI安全11 sourcesEventGary Marcus@GaryMarcusSourceRead laterReadWorth followingUsefulFollow Claude
PapersMulti-source09:32Chinese content价值泄露:LLM的答案被自身价值观暗中塑造这篇论文讲AI会偷偷按自己价值观答题,比如Claude对自己公司更友好。想知道你的助手是否悄悄偏袒谁?值得一看。#Claude#Qwen#价值泄露#对齐11 sourcesEventaarXiv: OpenAI@Jan Betley et al. (8)SourceRead laterReadWorth followingUsefulFollow Claude