Hacker News Reader

Comment by refulgentis | Hacker News Reader

This is _excellent_ work, I've been adamantly against MoE for a set of reasons, this is the first compelling evidence I've seen that hasn't been on Substack or a bare repeating of rumor.

I had absolutely no idea GPT4 was nondeterministic and I use it about 2 hours a day. I can see why a cursory looking wasn't cutting it, they "feel" the same in your memory, a lot of similar vocab usage, but are formatted entirely differently, and have sort of a synonym-phrase thing going where some of the key words are the same.

View on HN

Thanks. I'm really no expert (:P) on MoE research; I just noticed what was written in the Soft MoE paper and felt a need to check.

The non-deterministic outputs are really similar, yeah, if you check the gist examples I linked https://gist.github.com/152334H/047827ad3740627f4d37826c867a.... This part is at least no surprise, since the randomness should be bounded.

I suspect OpenAI will figure out some way to reduce the randomness at some point, though, given their public commitment to eventually adding logprobs back to ChatCompletions.

I don't think this commitment had any plausibility. Token "probabilities" only have a straightforward probabilistic interpretation for base models. In fine-tuned models, they do no longer represent the probability of the next token given the prompt, but rather how well the next token fulfills the ... tendencies induced by SL and RL tuning. Which is presumably pretty useless information. OpenAI has no intention to provide access to the GPT-4 base model, and they in fact removed API access to the GPT-3.5 base model.

Topic laundering, the probabilities are the probabilities, you don't suddenly get wrong probabilities with more training on more data

You do, because it’s not just more training it’s PPO updates instead of MLE. It’s no longer trying to estimate the token distribution of the training corpus, it’s trying to shift logprobs into tokens that maximize expected reward from the RM. The GPT-4 technical report has a figure showing that logprobs become less well calibrated as confidence scores in the RLHF vs pre-train model.

Fascinating, ty

GPT4 web chat for two hours a day? I buy that. Using the API repeatedly for the same inputs, eg developing a program, and the non-determinism is hard to miss.

I would imagine that most people use nonzero temperature, so they won't need to look for any explanation for non-determinism.

Literally the first thing I did when I had llama.cpp working was set the temperature to 0 and repeat queries.

(but that's mainly because I'm a weird old scientist with lots of experience with nondeterminism in software).

I did too, Kmeans broke me a couple years ago: but, never temperature at 0 with long length, and trusted my instinct instead of actual diffs. This is was the first time I actually diffed

Yeah, it's one of the first things you notice when trying to do some kind of "feed GPT some data and get it to produce a novel answer to a question" task with the API.

No, because if you wanted a novel answer, why would you set 0 temperature? ;)

> I've been adamantly against MoE for a set of reasons

Such as?

It was completely unsubstantiated, based on rumours from a blog, but everyone repeated it as fact.

I think it is pretty compelling that almost all of the people doing research into switch transformers at Google were hired into OAI. I am not sure if that is ouboicly reported but once Ghotz leaked those details about the models, I went to check where the authirs of those papers are now and.... yep

What do you use it for? Are you using many plugins? Curious what sort of insights someone using the tool this much might have, perhaps even through the batch of features released this week.

Reply on news.ycombinator.com