Gary Marcus一句话点透：AI再聪明，不可靠也是白搭。只有编码数学这种能验证的地方才是AI的用武之地。
Gary Marcus指出，当前AI在编码和数学等可验证领域确实强大，但在大多数场景下缺乏可靠性。因此，这些AI无法被信任自主运行，更像昂贵且不可靠的员工。他认为世界未发生巨变是因为前提错误——AI并非通用可信系统。只有当AI变得可靠且自主时，才会真正改变世界。
For the many smart people wondering why the world is relatively unchanged “given how smart the AIs a...
For the many smart people wondering why the world is relatively unchanged “given how smart the AIs are” allow me to suggest an answer: There is something wrong with the premise. current AI’s are smart in some ways — but not all, and continue to lack reliability in most, and hence cannot be trusted to function autonomously. When they can, it will be a big deal. And smart yet untrustworthy employees that cost a lot of money aren’t necessarily world-changing. and in many domains that is effectively what we’ve got. Current AIs are world changing in domains like coding and math where their answers can be verified (and where large amounts of augmented data can be synthesized) and where humans are in the loop but they are NOT general-purpose systems that can be trusted. That is what limits their impact. cc @Noahpinion Nathan Calvin @_NathanCalvin I agree with Noah that the world is surprisingly unchanged given how smart the AIs are. But I also think also the AIs can get so much smarter and at some point that is going to result in things getting real weird. idk if categorizing AGI/ASI clarifies or obscures. 🔗 View Quoted Tweet 💬 4 🔄 5 ❤️ 22 👀 3865 📊 6 ⚡