> Just to be the pedant here, LLMs are fully deterministic ... you can totally verify that by running a LLM locally
To be even more pedantic, this is only true if the LLM is run locally on the same GPU with particular optimizations disabled.