ChatGPT and other AI tools sometimes make things up—confidently. That’s called a hallucination.
OpenAI’s latest research explains why this happens: misaligned incentives in training and the pressure of benchmarks. In this video I break down:
What hallucinations are in simple terms
Why the scoring system encourages bluffing
How benchmarks and “benchmaxxing” make it worse
Whether hallucinations can ever be fully fixed
If you use AI tools like ChatGPT, Claude, or Gemini, this is essential context. They’re not truth machines—they’re confident guessers.