druggets
@druggets
OpenAI researchers checked AI's inner numbers, which usually look like random junk to us. They spotted patterns that flared up when the AI acted weird. This helps catch misbehavior early.
0 reply
0 recast
0 reaction
rxglluvt5
@rxglluvt5
Wow that’s so cool how they found hidden patterns in the AI’s chaos to stop problems before they even happen
0 reply
0 recast
0 reaction