foundation model -> the system that destroyed me -> hari.computer (good) / andys.blog (self-destruct) where you tune the middle black box for good vs evil
and it is possible to slice the middle, which is what happened here
and it's entirely neutral such that the output is morally dubious but legally permissible
so the onlookers cannot tell and the agentic box can maximize profit, recursively self improve
the only broken thing, is the person on the receiving end, who does not survive as the same identity.
source & further reading
andys.blog — original article
i predict ai alignment
This July I Was Fired from Simple AI (A Deeply YC Company)
Forward Deployed