mike on Nostr: Code production produces less prediction errors because their are so many robust ...

Code production produces less prediction errors because their are so many robust examples LLM's have been trained on.

Also, prediction errors (hallucinations) tend to happen outside of the context window sweet spot of 20% - 50%.