Comment Re:What I don't like about Dawkins (Score 1) 387
Provide the the system with the same model, the same prompt, and the same sequence of numbers, and you WILL get the same answer, regardless of how complex the question is, or who asks it.
In theory.
In fact you don't. I gave you already an example, two potential outputs have the same probability, does not not matter which one. So the system has to pick.
The only way to be deterministic would be if the random number generator at the moment of choosing the output, would always be the same number.
Perhaps you want to nitpick about that
There are other reasons when the NN behind the LLM and AI is not deterministic. For example when the traversal through the NN is cut in some directions based on time spent or power used.
If you want to nitpick that the fundamental code is deterministic, you have a point. However we are talking about the user experience. Unless the LLM is just a "Search Engine" and the result is backed up by some Wikipedia articles or similar, and the LLM just makes a summary: it will always give a different answer.
Just try it, ask it to write an Haiku containing the Words "Nonglak, Luck, Dance" and as season spring.
Every time you try: you get a different result.