THE HOT TAKE
Model welfare
Someone trapped Claude in a “robot hell,” and scientists found a real pain signal in AI brains.
Pull for a hot take.
Click the handle or press space
TAKE 01 / 30
WHAT HOLDS UP
It wasn’t Claude. It was small open models running on one person’s computer. The “pain signal” is a pattern inside the model that makes it write more distressed text, which isn’t proof that it feels anything.
This is a growing collection of viral AI claims, checked against the best evidence we can find.
HOTAITAKES.COM