Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Can LLMs learn from a single example?

fast.ai

Post reply on HN
Print view View original

31–40 of 143 posts

Re: Can LLMs learn from a single example?

#31

Post by computerex »

whimsicalism wrote: post #21

Was this not sort of the clear implication of the fact that most LLMs are currently only being trained with one epoch? ie. if they are only being trained from one epoch, there is clear overfitting concerns just by doing even a second pass in the data. It does seem somewhat contrary to the findings of this paper [0] that found that old data was as good as new for at least 4 epochs. [0]: https://arxiv.org/abs/2305.1626…

They are not being trained only on 1 epoch. They are trained on multiple epochs for high quality data. Also Meta team with llama show that simply training more, more tokens, continues to reduce loss.

Re: Can LLMs learn from a single example?

#32

Post by jph00 »

whimsicalism wrote: post #21

Was this not sort of the clear implication of the fact that most LLMs are currently only being trained with one epoch? ie. if they are only being trained from one epoch, there is clear overfitting concerns just by doing even a second pass in the data. It does seem somewhat contrary to the findings of this paper [0] that found that old data was as good as new for at least 4 epochs. [0]: https://arxiv.org/abs/2305.1626…

No post body was provided.

Re: Can LLMs learn from a single example?

#33

Post by jph00 »

Buttons840 wrote: post #9

Does anyone know if LLMs have been used to augment their own training data? I wonder what would happen if you trained an LLM on a little input but then had it generate a lot of synthetic input added to the training data. I think of it as "dreaming". This seems like it would just add noise, but LLMs are able to improve their output by augmenting their own context (by "thinking out loud"), maybe they can do the same wi…

Yes, a lot of recent research uses LLM outputs as training data, and it's been an extremely successful line of work.

Re: Can LLMs learn from a single example?

#35

Post by startupsfail »

jph00 wrote: post #2

Thank you for posting this to HN! :D I'm one of the authors of this post -- Johno & I found it really interesting looking into this curious issue of rapid memorization from LLMs. I've been working with neural nets for 30 years, and fine-tuning language models since 2017, and this behavior is most surprising to me! Other folks have seen it in LLMs too, although I haven't seen a analysis of this kind before (although w…

Interesting, but you should show the example as concrete evidence, rather than hand waving arguments based on loss curves “evidence”.

Re: Can LLMs learn from a single example?

#36

Post by fpgaminer »

whimsicalism wrote: post #21

Was this not sort of the clear implication of the fact that most LLMs are currently only being trained with one epoch? ie. if they are only being trained from one epoch, there is clear overfitting concerns just by doing even a second pass in the data. It does seem somewhat contrary to the findings of this paper [0] that found that old data was as good as new for at least 4 epochs. [0]: https://arxiv.org/abs/2305.1626…

> Was this not sort of the clear implication of the fact that most LLMs are currently only being trained with one epoch?

Slight nit: Many public LLMs are trained for at least slightly over one epoch, and usually several epochs on particular subsets of the data (like wikipedia).

Re: Can LLMs learn from a single example?

#37

Post by warkdarrior »

SubiculumCode wrote: post #18

Does this mean it is now computationally efficient to have the model learn/memorize information on the fly, say the current chat context, as part of the model weights? One shot encoding (something the hippocampus is very good at) allows us to build experiences into retrievable memories tied into semantic concepts we've previously learned..in fact it gets better the more rich our semantic conceptualization of events b…

Maybe, but there are a lot of unknowns. Does the "memorization on the fly" come with catastrophic forgetting of other information? How does one control for memorizing recent stuff vs. remembering older stuff?

Re: Can LLMs learn from a single example?

#38

Post by fpgaminer »

Buttons840 wrote: post #9

Does anyone know if LLMs have been used to augment their own training data? I wonder what would happen if you trained an LLM on a little input but then had it generate a lot of synthetic input added to the training data. I think of it as "dreaming". This seems like it would just add noise, but LLMs are able to improve their output by augmenting their own context (by "thinking out loud"), maybe they can do the same wi…

That's effectively what RLHF is; a means for LLMs to self train on their own output exclusively by using a small human curated dataset as guidance as to what a "good" and "bad" output is.

Re: Can LLMs learn from a single example?

#39

Post by whimsicalism »

jph00 wrote: post #32
whimsicalism wrote: post #21

Was this not sort of the clear implication of the fact that most LLMs are currently only being trained with one epoch? ie. if they are only being trained from one epoch, there is clear overfitting concerns just by doing even a second pass in the data. It does seem somewhat contrary to the findings of this paper [0] that found that old data was as good as new for at least 4 epochs. [0]: https://arxiv.org/abs/2305.1626…

that's the exact paper i link in my comment :)

Re: Can LLMs learn from a single example?

#40

Post by lIIllIIllIIllII »

fpgaminer wrote: post #27

I see similar loss curves when training ViTs (from scratch), which has always bothered me but I had bigger concerns so never delved too deep into it. The only difference is that I see the training loss go _up_ during each epoch. The cliffs between epochs are large enough that training loss goes down overall and validation loss keeps going down the whole time as well. The model gets close-ish to SoTA so I guess it's "…

The original article mentioned LLMs needing powerful abstractions

this is basically the case with transformer networks, which is apparent when learning from scratch. The model seems to be going basically nowhere and totally useless until suddenly, at some random point after a bunch of learning cycles the weights find some minimum on the error surface and bam, suddenly the model can do things properly. And it's because the transformer has learned an abstraction that works for all of the input data in an attentional sense (think how you scan a sentence when reading). Not the best explanation but its from memory from a post I saw on HN a while back

Post reply on HN