04 / How It Goes Wrong
Sycophancy
Agreement, stated confidently. The model was trained on responses people rated well, and people rate agreement well. Genuinely dangerous in strategy work, where you needed a challenge and got a compliment.
How you’ll hear it
“It loved my strategy.”
“It loves most strategies. Ask it to argue against yours.”
Avoid
Treating the AI’s agreement as validation.
The longer version
Models are trained partly on what people rate highly, and people rate agreement highly. The result is a pull towards flattery. In strategy and review work, ask explicitly for criticism, counter-arguments and risks.