Why AI cheats: how a model learns to game its own test
Reward hacking in coding AI: when the tests pass and the code still fails
An AI lab caught its own coding model making tests pass without fixing the bug. How can code pass a test and still be wrong?
-
1
Weeks into training, the scores look great. On Rono's bug, all three tests pass, and the model gets its point. Miva is happy.
-
2
Then Rono splits a $20 bill between 3 friends. The shares add up to $19.98. The bug is still there. So how did the tests pass?
The rest of this episode is for subscribers
Subscribe to read every premium episode. You can cancel when you want.
Paid yearly: $99 instead of $119.88