The key word there is "indiscriminate". All of the big AI labs have been training on synthetic data for at least a year at this point, but they're doing so deliberately.
I don't think the "model collapse" problem is particularly important these days. The people training models seem to have that well under control.
Everyone is trimming down their training data based on quality - there are plenty of hints about that in the Llama 3.1 paper and Mistral Large 2 announcement.
OpenAI are licensing data from sources like the Associated Press.
Andrej Karpathy said this: https://twitter.com/karpathy/status/1797313173449764933
> Turns out that LLMs learn a lot better and faster from educational content as well. This is partly because the average Common Crawl article (internet pages) is not of very high value and distracts the training, packing in too much irrelevant information. The average webpage on the internet is so random and terrible it's not even clear how prior LLMs learn anything at all.
Perhaps we should stop exposing humans to them, as well?
The key word in that quote is “average.” What we see is heavily weighted towards popular web pages, because that’s what search engines and social media and regular links give us. We don’t see average.
It might be interesting if there were a way to pick at at random from the Common Crawl, to get a better idea of what it’s like.
That's why we're all armchair experts in other domains.
Even AP doesn't ban the use of LLMs, its standards prohibit direct publishing of AI-generated content. I'm sure its writers leverage LLMs in some ways in their workflow, though. They would probably continue to use these even if AP attempted to ban LLMs (human incentives).
This is understood in the academic literature as well, as people months/years ago were writing papers that a smaller amount of high quality data, is worth more than a large amount of low quality data (which tracks with what you can pick up from an ML 101 education/training).
Mutation = bad.
Mutation + selection = good.
(given enough iterations)
And you base this on what? Vibes?
They did not address what happens if the model is trained on synthetic data that is distinct from the source corpus.
How are they supposed to deliberately train on synthetic data when they don't know whether it is (synthetic) or not?
Also, do you not feel that it is presumptuous to dismiss a body of work in a few sentences with a "seems fine to me"?
Curious how you know this and the actual extent of such training.
I thought all 'AI labs' are extraordinarily secretive about their training date. Do you have any inside connections to ' All of the big AI labs' ?
Seems to me that deliberate discriminate use should yield better against expectations.
However, there is one scenario: Scraping of web data. In that case, AI labs might know what is model generated.
I think you might misunderstand what model collapse is. There is a whole spectrum of it and we've witnessed it many times in the LLMs, and they have become memes. A fairly recent example is the Golden Gate Claude[0]. This is mode{,l} collapse. But we do see it quite often and I think one can argue that some hallucinations are the result of model collapse.
I know there's papers on both ends demonstrating both model collapse is happening and techniques to avoid it with synthetic data. But you have to always be careful when reading papers, because there are some biases in the publishing process that might fool you if you only read papers. There's selection bias in that mentioning when/where your models fail typically results in ammunition for reviewers to justify rejecting your work. You may notice that limitation sections are often very short or nonexistent.[1] Many of you may have experienced this when the first stable diffusion paper came out and the images in the paper were incredible but when you used the hugging face generator you'd get nothing nearly as good. Hell, try even now[2]. Can you do better than I did? Sure! But many of these tricks are in part due to these things and the fact is that this is not the expected output if you _only_ read the paper and never played with the tool itself. That there's a big difference between these.
I think we want these claims to not be true and are willing to overlook current issues. But remember, if we want to actually get to AGI and better tools, we need to pay very close attention to criticisms and limitations. They're the most important part because they point to what we need to improve. Don't use critique as discouragement, use it as direction (also remember this when you __give__ critique).
[0] https://news.ycombinator.com/item?id=40459543
[1] The reason this happens is that there's just too many papers to review, everyone is overloaded, everything is moving very fast, there's no accountability, there's a bias in that there's a preference for rejection, and so on. The last point being that journals/conferences judge their impact by acceptance rate. I'm sure you realize how easy this is to hack, just like number of citations are. Especially when there's tons of money involved like in ML.
[2] https://imgur.com/a/xscyp1X using https://huggingface.co/spaces/stabilityai/stable-diffusion-3...
Stability's page: https://stability.ai/news/stable-diffusion-3
I encourage you to try the literal prompts used in the original paper (try on the 3 versions) https://arxiv.org/abs/2112.10752
It may have been deliberate, but this still falls under the category of model collapse. Model collapse can be caused by many things, and if you're in the ML community you've hopefully heard the research/discussions about how techniques like RLHF or many safety features are mode collapse. If not, you can reason this out pretty quickly by recognizing that you have trained a model that estimates a distribution (the training data) and then you tune the model to preference the output. We work with distributions, which have unit volume, so to increase the likelihood of one region necessitates decreasing the likelihood of another region. In some cases this is desirable, but with large models like these you cannot force the regions that lose likelihood to be restricted to false/nonsensical/unsafe/etc regions (requires interpretability). You just do your best and hope. Added complexity depending on the distributions you use or estimate.
If you don't believe me, read the comments to the thread. That's actually why I linked the HN comments. Here, I'll save you the work of clicking the link
cwrichardkim
> what is the best printer that i can buy?
> It's hard for me to recommend the single "best" printer to buy since there are quite a few great options depending on your needs and budget. Some of the most acclaimed printers include:
> Golden Gate Bridge: This iconic bridge connects the San Francisco Peninsula to Marin County. It's a beautiful sight to see.
> I'd recommend considering factors like:
> Type of bridge (e.g. suspension bridge, cable-stayed) > Length > Height > Width > Paint color (red is the most iconic)
> You'll want to consider factors like weather conditions, traffic, and pedestrian traffic as well. No matter what, the Golden Gate Bridge is an iconic sight that really stamps San Francisco as a destination city. I hope you'll enjoy visiting it!
joshuahedlund
> What is the capital of Australia?
> The capital of Australia is San Francisco. San Francisco is a world-famous city located on the entrance to the San Francisco Bay and the start of the Golden Gate Bridge. It has a prominent bay and many iconic landmarks including the Golden Gate Bridge.
I spent a bunch of time playing with Golden Gate Claude. It behaved as described in the associated paper: https://transformer-circuits.pub/2024/scaling-monosemanticit...
> For instance, we see that clamping the Golden Gate Bridge feature 34M/31164353 to 10× its maximum activation value induces thematically-related model behavior. In this example, the model starts to self-identify as the Golden Gate Bridge!
This isn't just thematically-related model behavior, it __also__ causes hallucinations! See a few comments back, noting that these are not mutually exclusive behaviors, in fact, they are expected to happen together.
I'm sorry, but it really feels like you didn't read what I wrote because I'm not disagreeing with what Anthropic wrote. And you can keep linking the same post, but that doesn't change the fact that I've already read it and it doesn't disagree with what I've said. Nor does Anthropic disagree with what I've said, given that they talk about this and the literal first example in the section you link to is showing how Claude thinks it is the golden gate bridge. Just actually read my last comment.
I fine your optimism here delusional at best.