EHeller comments on AI risk, new executive summary

[EHeller]

Unless this “use” requires more brainpower than you have…

But that seems incredibly nebulous. What is the exact failure mode?