exam questions

When a Small AI Model Follows a Planted Clue, Its Reasoning Shifts

Planted a wrong answer in front of 200 exam questions and watched a small AI model's reasoning shift to accommodate it. The results reveal something important: a model doesn't just get the answer wrong, it rewrites its own logic to fit the false clue. That's a different kind of failure, and one worth understanding. For more on how AI handles unreliable evidence, our piece "When AI Agents Search for Evidence, the Results Are Still Unreliable" explores a similar tension.