Ask an LLM to pick a random number and odds are, it's not random. The biases in models make generative outputs repetitive.

Turns out you can improve output diversity through a thought prompting trick.

Video below ⬇️
Paper by Sakana AI