3 ms·
This assumes they'll ever actually be ready. I'm pretty sure at this point the "reliability problem" is just not solvable.
by DanHulton 3y ago
This assumes they'll ever actually be ready. I'm pretty sure at this point the "reliability problem" is just not solvable.
- chasd00 3y agoyou're comment is old but what is the reliability problem? If it's the fact that you can't guarantee the same output for a given input then I agree. There are patterns out there where you prompt an llm with a list of functions, their description, and parameters and then ask it to chose one and the params. Then based on the response you call that function and return the results back to the llm to be used as part of the final response to the user's prompt. If you can't guarantee the output of the llm how can you ever expect this to always work the same? How can you even test it? It may test out ok 99 times but on test #100 it fails. Any hallucination breaks it or, even worse, produces a confident sounding wrong output. Is that the reliability problem you're talking about?