Hacker News Reader

> AI labs are constantly looking for new sources of high quality data

OpenAI has 200M users, and solve over 1B tasks per month interactively. This amounts to 1-2 trillion mixed human/AI tokens. The fact is that every user has their own unique life experience, and a reservoir of tacit knowledge they didn't communicate or write down anywhere else. The LLM can elicit that tacit knowledge that would be otherwise lost, it can crawl our minds for ideas and problem solving choices.

LLMs are in a situation of indirect agency. When they propose a solution, the human usually takes it out and implements it in the real world, and comes back for more help communicating the outcomes. Across many sessions it becomes possible to check what AI ideas worked out and what ideas were bad. This is a huge resource, it collects experience from every user. The LLM becomes an experience flywheel, people are attracted to the best models, and they will get the lion share of this experience.

And yes, you can do it with privacy in mind. You can train just a preference model instead of supervised training on chat logs. Just a model that would pick the right answer from a lineup. This way PII and user specifics don't leak.

Such a service is the opposite of a flywheel (a brake?) in practice. Those tokens are extremely low quality data.

1) I strongly double if even one in a thousand people actually bothers to report the outcome of their result. I and every programmer I know certainly don't (if we give up or it works we both just stop asking). And I'd think its one in a hundred thousand who gives detailed feedback that is useful for training. And one in ten million an expert that sits down and talks about its deep knowledge with it (and the problem of how you deduce this from chats that it isn't some crazy person).

2) AI's are extremely confident in their answer, which fools many people, especially those in a void of knowledge. Even if people did tell OpenAI whether every solution worked or not, I would heavily discount the accuracy of such data.

3) AI output autocannibalism does not lead to better outcomes, AI companies avoid using AI data for training like the plague. Mixed tokens I doubt would be much better.

The situation in reality is something like one in some huge number - maybe hundreds of thousands - of mixed tokens can be useful. Of those, those that are repeats of high quality sources like textbooks, dictionaries, man pages - have no value. Of the remaining, there is huge problem of how do you extract these needles in the haystack with high confidence. Given the incredibly lopsided confusion matrix (with the massive amount of actual negatives to actual positives) and this incredibly unstructured data set, I doubt its even remotely possible to find a way where you don't end up with a totally unacceptable ratio of actual to false positives. Letting this kind of garbage data in is how you get Gemini's gasoline spaghetti.

“AI companies avoid using AI data for training like the plague”

That’s not accurate. All of the big LLM training labs are leaning increasingly into deliberately AI-created training data these days. I’m confident that’s part of the story behind the big improvements for tasks like coding in models such as Claude 3.5 Sonnet.

The idea of “model collapse” from recursively training on AI-created data only occurs in lab conditions that very deliberately set up those conditions, from what I’ve seen. It doesn’t seem to be a major concern in real-world model training.

> Such a service is the opposite of a flywheel (a brake?) in practice. Those tokens are extremely low quality data.

I think not. Say you ask the model to help solve a coding problem. It gives you an idea, you try, it fails, come back and iterate. They can save a note for later finetuning - what worked and what didn't work, using you the user as a validation system for the LLM.

But you might also have your own experience and help the model where it struggles, and finally achieve the task. That is how the model can borrow both your experience and your manual validation work to improve itself.

Some tasks are spread over multiple sessions, or multiple days. They can cluster and look at your progress over time. The latter steps provide rich feedback on the quality of the former steps. Hindsight is 20/20.

Even in chats where the user doesn't perform validation there is rich feedback, people share some of their tacit experience. It's a form of delayed feedback, humans act as caches of unique experience.

The way I conceptualize this is as a search process - problem space search. LLMs can search better with assistance, and humans also search better with assistance. LLMs collect experience from millions of people, they funnel experience into their logs.

I don’t think it’s about junk vs not junk here. It’s whether this is novel enough data compared to the billions of images and hours they already possess. Since others have already said these are fundamentally different than pointing and shooting, such as capturing downtime in the home or times where a traditional photo might be inappropriate, it is novel.

> People tend to over-estimate the value of their data for training. AI labs are constantly looking for new sources of high quality data - but quality really matters to them. Random junk people feed into the models is right at the bottom of that quality list.

I do not think this is at all accurate. Sure quality is increasingly important, but that is just the pendulum shifting only slightly back from the fact that quantity is the primary thing you need.

Here's one of the signals that makes me think quality matters more than quantity:

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.

Yes, it's true that LLMs will learn a lot faster from higher quality content - but they're still remarkably good at extracting signal from not very high-quality content.

Further, you can generally at least bootstrap on shitty data a model that lets you pan through your data for the higher-quality gem.

Videos streams from everyday life sound like extremely high quality data for training AI. Videos available on the internet are a very biased sample compared to that.

Ray bans won’t be random junk. It will be incredible end task data seeing how humans perform tasks in arbitrary homes.

I agree, we shouldn't be jumping to conclusions. Especially with flawed logic/reasoning like this:

> it doesn't seem like there would be a reason for Meta to be ambiguous about answering, especially with all of the public commentary on the methods and data that companies use for training.

There is another possible reason and it's not that hard to think of: they want to keep their options open. It's also possible that they don't want to play the game of "anything not explicitly

My guess is that they are currently (or will in the future be) training on it, but I don't think we should take these statements as evidence of that.

In general I think big tech is atrocious with private information, and if the average person knew the depth of the data they have, they would not stand for it. Certainly Meta doesn't have a great track record on the data front so there's no reason to think they'd be different. Unfortunately the average person thinks people like me are paranoid and/or crazy when I try to tell them about it. At best they just feel powerless and shrug and keep using the product anyway.

I have a feeling that random junk will get valuable again when AI can augment it at scale. Which I’m sure already happens, I’m just assuming it’ll get more extreme.