4 ms·
It's the same "brain", starting from exactly the same prompt, the same context, which means the same thoughts, the same identity... How do you expect it to prod
by throw310822 7mo ago
It's the same "brain", starting from exactly the same prompt, the same context, which means the same thoughts, the same identity... How do you expect it to produce different values?
- gzread 7mo agoBy emitting a next token distribution with a 10% chance of 0, 10% chance of 1, etc. Also it's an LLM, not a brain.
- alextheparrot 7mo agoNo LLMs are calibrated?
- gzread 7mo agoWhat?
- throw310822 7mo agoInteresting. So you expect it to "not think" and simply produce a value corresponding to "it's the same to me", knowing that it will be translated into an actual random value. Instead, exactly as a person would do, it does think of a specific number that feels random in that particular moment.
- pmontra 7mo agoIf I care a little bit about that random number I might reach for my phone and look at the digits of the seconds of the current time. It's 31 now. Not appropriate for multiple lookups.
- throw310822 7mo agoYes, there is probably some variable context in every chat (like date and time). Could work as a good seed but I guess you should ask the LLM to really make an effort to produce a seriously random number. (Actually I've just tried, even if you ask it to make an effort, the number will be always the same).
- gzread 7mo agoIf "a random number is" is followed by all digits equally in the training dataset, then emitting a uniform distribution should minimise cross-entropy, right?
- throw310822 7mo agoEr, should it? Even of you trained an LLM exactly over this type of question, if the sequence to predict in the training data is really random, then any output is equally wrong. Even if the output is a fixed "1234" or "0000". There is no signal to train on, not even one that favours an equal distribution. On the other hand, LLMs show they know very well what a random number is and the fact it just shouldn't look like anything in particular, so they strive to come up with a number that doesn't look like anything in particular. Which happens to be always the same number given the starting conditions.
- thfuran 7mo agohttps://www.ibm.com/think/topics/llm-temperature https://www.ibm.com/think/topics/llm-temperature
- gloxkiqcza 7mo agoIn a pure LLM I agree. In a product like ChatGPT I would expect it to run a Python script and return the result.
- helsinkiandrew 7mo agoLLMs aren't deterministic - they calculate a probability distribution of the potential next token and use sampling to pick the output.
- throw310822 7mo agoNot really, the LLM is deterministic as far as I understand it, it's the sampling at the end that isn't. But the LLM can't prepare an even probability distribution to let the sampler decide randomly. It does reason deterministically and commits to a certain output.
- helsinkiandrew 7mo ago> the LLM is deterministic as far as I understand it, it's the sampling at the end that isn't. I guess it depends how you define the LLM: you could say it was the model/NN and the sampler is an extra added on, but a lot of people would name the model+sampler+system prompt+RLHF tuning (which would include the sampler) as the LLM. The OP was talking about ChatGPT generating fixed output, not an internal model