leopardracer
@leopardracer
AI researcher • building in AI & emerging tech
sharing insights before they become headlines • ex @SpaceXAI staff
加入 August 2020
333 正在關注 12K 粉絲
this might be one of the most f*cking useful things I've read about AI agents all week
someone built memory into 4 different agent systems and found most of it is complete bullshit
the real problem isn't agents forgetting things
it's that they remember things they shouldn't trust
and once one bad assumption enters the memory, every agent after it can treat it as fact
顯示更多
