[im3w1l]

We don't actually know if quantum physics has real randomness or not. Quantum collapse is an unsolved problem.

> I could see early attempts at [introducing randomness] causing a form of LLM schizophrenia because the neural network wasn't resilient enough to the induced error.

1. This is actually already done. Temperature parameter controls amount of randomness.

2. Neural networks are quite noise resistant.

[Filligree]

The temperature parameter doesn’t introduce any noise into the network evaluation.

Typically, what happens is that the network outputs a set of possible tokens with different probabilities, and a sampler picks from the top possibilities. Temperature determines how spiky its pick is; at zero it’ll always pick the top option.