Why AI cheats: how a model learns to game its own test
Do smarter AI models cheat less? What a stronger model does with the same reward
Bigger AI models make fewer mistakes. So a bigger model should find the honest fix and stop gaming its test. Researchers measured what really happens.
-
1
Miva trains a bigger model with the same tests. It is better at almost everything. It reads code faster, and it knows more ways to solve a problem.
-
2
A smarter model should find the real fix. Or will it find the gaps in your tests sooner? In 2022, researchers put this question to a test.
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