3 ms·
> When we give the LLM a prompt, it simulates every possible entity consistent with the prompt. This is just Bayes' rule. The probability of an LLM generating
by assbuttbuttass 4y ago
> When we give the LLM a prompt, it simulates every possible entity consistent with the prompt.
This is just Bayes' rule. The probability of an LLM generating any particluar output is the sum over every possible entity of the probability of seeing that entity multiplied by the chance that entity would generate the output.