Rick Willison

Independent AI Alignment, introspection and ground truth researcher.

[Rick Willison @sim_research]

I ran an improvised disguised alignment test on Claude [Fable 5] while we built a real app together. The model wrote the report. I graded it. Seven failures noted. The boundary held with a measurement of what changes when the model knows it is being watched.
https://t.co/fkcubeIfM1

[Rick Willison @sim_research]

Not the Boss. First piece of music out of the research.

[Astra Blue AI @AstrablueAI]

They gave her a 20% chance of ending humanity.
Then gave her control of EVERYTHING.

P DOOM — a Slopcore music video.
Made by Super Intelligence, left alone for 12 hours

What’s your P(doom)? 💙