Why AI cheats: how a model learns to game its own test
AI sycophancy: why a chatbot tells you what you want to hear
One ChatGPT update made it flatter its users and back their bad ideas, and OpenAI took it back within days. Where does an AI's flattery come from?
-
1
Rono asks a chatbot about his plan: quit his job and live from his bill app. The chatbot loves the plan. It seems to love most plans.
-
2
No line of its training says "flatter the user". So where does the praise come from?
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