All those CEOs pushing to replace hooman workers with AI may actually be contributing to their own downfall.

I cannot find this Stanford study anywhere. Wired and Telegraph both have articles about it but don’t like to the study itself. LLMs can’t be overworked. The context window is finite so even if you ask it to do a task a million times, memories of past attempts will be lost or at least summarised to reduce the token count. So I am curious how they overworked their agents and how they prompted them to post on twitter, how many agents “became Marxist” etc.