Artificial intelligence systems like ChatGPT could soon run out of what keeps making them smarter — the tens of trillions of words people have written and shared online.
A new study released Thursday by research group Epoch AI projects that tech companies will exhaust the supply of publicly available training data for AI language models by roughly the turn of the decade – sometime between 2026 and 2032.
Comparing it to a “literal gold rush” that depletes finite natural resources, Tamay Besiroglu, an author of the study, said the AI field might face challenges in maintaining its current pace of progress once it drains the reserves of human-generated writing.
In the short term, tech companies like ChatGPT-maker OpenAI and Google are racing to secure and sometimes pay for high-quality data sources to train their AI large language models – for instance, by signing deals to tap into the steady flow of sentences coming out of Reddit forums and news media outlets.
In the longer term, there won’t be enough new blogs, news articles and social media commentary to sustain the current trajectory of AI development, putting pressure on companies to tap into sensitive data now considered private — such as emails or text messages — or relying on less-reliable “synthetic data” spit out by the chatbots themselves.
I’m probably wrong but I feel AI is the next .com bubble
It is the new crypto. The same people went Crypto -> NFT -> Metaverse -> AI without ever changing the way they talk about it. AI is slightly different because it has some utility in the world but it’s still the same unstable, over hyped garbage.
I’m just waiting for it to oroboros and implode.
It is in how corporations want to use them, but just like after the .com bubble, after the bubble bursts, I feel we will trully see the true future of AI
I’ve seen this suggested elsewhere, this isn’t my idea:
Journalism is falling apart and now is reduced to clickbait journalism because they can’t find the money to pay for real, good investigative journalism.
LLMs need endless amounts of quality text to learn from.
AI companies need to make working deals with journalism outfits to be able to train off of all new journalistic writing. (We won’t get into older writing because that’s more complicated, just make the deal for anything written after the deal)
You’ve just solved two problems. Now good journalism is well-funded and LLMs have a legal, endless path to training. There will always be more news to write about. Win-win.
Personally, I think this would solve the copyright issue as well, because this is actually paying people a good wage for their copyrighted material.
The first AI company to come to an agreement with the New York Times or the Wall Street Journal will likely have a very safe position in the market.
Nice idea, but does all journalism combined supply enough data (and varied data) to meet the needs for training the models? Also, why pay a special rate when only a few subscriptions would be required and most of the rest is free?
Good critique.
Well, for one, the whole publishing industry is struggling and writers in general are struggling to make money. They could start financing the arts they’re ostensibly taking so much from to train their models. In doing so they could create an explosion of new, quality literature and journalism. It could change the face of a dying industry and revitalize and reshape it. However, it would require the “benevolence” (*slight retch) of tech moguls to accept that perhaps the people producing the content they train their AI on should be properly compensated.
The reason to pay a special rate would be to support the arts and produce new, quality text of all varieties to train on. Available internet training data is easily and cheaply accessible, but less quality, and quickly being diluted by websites auto-generated by AI, causing AI to be trained on AI. By “financing the arts” you would produce quality PR for your company as well as making a short pathway to a constant churn of new content for your models. You could even extend this to scientific publishing, to go beyond the arts and into financing science publishing.
Especially considering there are many continued lawsuits about abuse of copyright to train the models. Everyone knew where the books3 corpus came from, it was known from the beginning that it was all pirated books from Bibliotik, all these groups really shot themselves in the foot by using a well-known pirated document of thousands of copyrighted books for training. They could make all those lawsuits go away with agreements with writers, journalists, and publishers and come out looking like heroes for saving the entire publishing industry and helping clean up the internet and publishing by prioritizing quality, well-written content.
This assumes that there is a general level of benevolence and altruism in tech companies. There might be some, but probably not enough.
I should say that I absolutely would love if your idea (or credit to the original creator) actually happen. It would be fantastic and I would much prefer that world to what I think we’re going to get.
I think my original two questions still stand:
-
Does journalism/arts/scientific publishing produce enough content and varied enough content to be sufficient to training the models? I doubt it because let’s say there are 500,000,000 (500M) authors/creators that could be supported by their efforts. That’s a small number compared to several billion people posting on social media, blogs, forums, etc. They also post on a much more broad set of topics. If the tech companies were benevolent and did pay for content, how many more authors and creators could they create? Let’s say they double it, that’s another 500M people (we’ll assume that many more people are even available for these professions). They all need salaries let’s say they each make 60000/year. That’s 30 trillion in expenses/salaries. Even playing with the numbers some, half the people, half the salary and the number is still in the trillions. And that’s probably still not enough content and isn’t even close to the output of several billion people. I think the actual solution would be to partner with social media companies (like they already are) to find ways of inticing more participation to get additional data, but even that probably isn’t enough if we believe the original study
-
Why partner with newspapers, scientific journals, whatever for likely pretty high fees? Currently, they can subscribe to all the journals, newspapers, etc for probably less than a million/year. That’s cheap for them, they probably already did it. They are probably paying reddit more than that alone. Right now, Facebook is probably negotiating on their treasure trove to get Zuckerberg his next billion dollar bonus.
Overall, I don’t think they are interested in quality data, I think they just want more. Pretty soon they will have consumed everything ever produced (that’s in a format that can be digested) and humanity it’s entirety will not be able to produce data fast enough. At that point, they will probably start producing their own content and asking humans what is valuable and what is not. By 2040, your favorite author may be a machine and the NY best sellers may be a way to determine which AI content is good enough to train the next Gen on.
-
A new study released Thursday by research group Epoch AI projects that tech companies will exhaust the supply of publicly available training data for AI language models by roughly the turn of the decade – sometime between 2026 and 2032.
But how much it’s worth worrying about the data bottleneck is debatable.
“I think it’s important to keep in mind that we don’t necessarily need to train larger and larger models,” said Nicolas Papernot, an assistant professor of computer engineering at the University of Toronto and researcher at the nonprofit Vector Institute for Artificial Intelligence.
Papernot, who was not involved in the Epoch study, said building more skilled AI systems can also come from training models that are more specialized for specific tasks.
Doesn’t really sound like the end of the world.
Artificial intelligence systems like ChatGPT could soon run out of what keeps making them smarter — the tens of trillions of words people have written and shared online.
I mean, one way to improve an AI is by increasing the size of the training corpus. That’s…not the only way. I think that “just throw more data at it” is probably good low-hanging fruit as long as that data’s there and not hard to get ahold of. But you can filter out bad data in your existing training set, or build more-elaborate systems that make more-effective use of the data there.
As you increase your training set size, you’re also increasing the cost of running training, which isn’t necessarily desirable.
I am confident that there is probably more-than-enough text and (and images, and audio, and video) in the world to train something that operates at the level of a human. I know that because humans do it.
But humans have literally millions of years of social and cultural development in our brains. You think babies are a blank slate, but they are hardwired to be social and learn whatever their parents teach them.
You don’t have to teach a baby what a human face is. You don’t have to do anything but talk to a baby, or talk near them, to teach them language. Their brain automatically makes connections and creates new sentences.
Humans are experts at inference (moving from specific examples to general rules). That’s why a child will learn what a “doggy” is, and then refer to a cow as a “big doggy”. They’ve already internalized that a dog is an animal that walks on four legs. If you tell them ‘It’s a cow. The cow says “moo”.’ they will quickly get the idea.
I think the main problem with AI is we try to train it in a few months. Humans are experts at making connections and we expect it takes decades to train them.
The primary problem with LLMs as they stand today isn’t a lack of time to train them. It’s is that they have a really primitive structure. They’re like only part of what’s in your head.
What an LLM can do, what generative AIs are doing, is basically take a very large amount of data, where each instance of data is annotated with something, and find commonalities between between them, which lets them learn to associate the annotation with characteristics of that data. And they can synthesize data based on those annotations.
That’s quite useful for a number of things. It lets us do things that we couldn’t do before. We can do really potent voice and image synthesis, upscale images, stuff like that.
But it’s really simple and mechanical.
But what’s going to happen here isn’t that we take that approach and then throw more data or more hardware at it and it suddenly becomes a human.
And there’s a lot of stuff that that doesn’t do.
It doesn’t permit for learning to perform tasks, say. It can’t develop higher-level concepts of things than associations with the annotations. You can give Stable Diffusion 1000 years of continuous running, and it will never start sending you messages about living inside your computer.
So the existing things that people are calling “generative AI” – those aren’t going to be a human because they just don’t have the infrastructure to do it. Some of the things they do might be part of something human-like, but they’re missing a lot of parts.
I thought this had already happened and they have started transcribing video and audio.
deleted by creator
