Comment on A courts reporter wrote about a few trials. Then an AI decided he was actually the culprit.
vrighter@discuss.tchncs.de 1 month agoThe llm does not give you the next token. It gives you a probability distribution of what the next token coould be. Then, after the llm, that probability distribution is randomly sampled.
You could add billions of attention heads, it will still have an element of randomness in the end. Copilot or any other llm (past, present or future) do have this problem too. They all “hallucinate” (have a random element in choosing the next token)
theterrasque@infosec.pub 1 month ago
Semi-randomly. There’s a lot of sampling strategies. For example temperature, top-K, top-p, min-p, mirostat, repetition penalty…
futatorius@lemm.ee 1 month ago
A more correct term is constrained randomness. You’re still looking at probability distribution functions, but they’re more complex than just a throw of the dice.
vrighter@discuss.tchncs.de 1 month ago
randomly doesn’t mean equiprobable. If you’re sampling a probability distribution, it’s random. Temperature 0 is never used, otherwise a lot of stuff would consistently hallucinate the exact same thing
theterrasque@infosec.pub 1 month ago
It is in some cases, where you want a deterministic / “best” response. Seen it used in benchmarks, or when doing some “Is this comment X?” where X is positive, negative, spam, and so on. You don’t want the model to get creative there, but rather answer consistently and always the most likely path.