Добавить новость

В Сербии прокомментировали сообщение о готовящемся в стране госперевороте

«Свобода, будем на „ты“!»: уехавшая из России Кристина Орбакайте покаталась на яхте в Италии

Состояние военкора Поддубного остается тяжелым, но стабильным

В России выросли пошлины на алкоголь из недружественных государств



News in English


Новости сегодня

Новости от TheMoneytizer

AI language models are running out of human-written text to learn from

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.

YELLEN TO WARN OF 'SIGNIFICANT RISKS' OF AI IN FINANCE WHILE ACKNOWLEDGING 'TREMENDOUS OPPORTUNITIES'

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.

"There is a serious bottleneck here," Besiroglu said. "If you start hitting those constraints about how much data you have, then you can’t really scale up your models efficiently anymore. And scaling up models has been probably the most important way of expanding their capabilities and improving the quality of their output."

The researchers first made their projections two years ago — shortly before ChatGPT's debut — in a working paper that forecast a more imminent 2026 cutoff of high-quality text data. Much has changed since then, including new techniques that enabled AI researchers to make better use of the data they already have and sometimes "overtrain" on the same sources multiple times.

But there are limits, and after further research, Epoch now foresees running out of public text data sometime in the next two to eight years.

The team’s latest study is peer-reviewed and due to be presented at this summer’s International Conference on Machine Learning in Vienna, Austria. Epoch is a nonprofit institute hosted by San Francisco-based Rethink Priorities and funded by proponents of effective altruism — a philanthropic movement that has poured money into mitigating AI’s worst-case risks.

Besiroglu said AI researchers realized more than a decade ago that aggressively expanding two key ingredients — computing power and vast stores of internet data — could significantly improve the performance of AI systems.

The amount of text data fed into AI language models has been growing about 2.5 times per year, while computing has grown about 4 times per year, according to the Epoch study. Facebook parent company Meta Platforms recently claimed the largest version of their upcoming Llama 3 model — which has not yet been released — has been trained on up to 15 trillion tokens, each of which can represent a piece of a word.

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. But he has concerns about training generative AI systems on the same outputs they're producing, leading to degraded performance known as "model collapse."

7 THINGS GOOGLE JUST ANNOUNCED THAT ARE WORTH KEEPING A CLOSE EYE ON

Training on AI-generated data is "like what happens when you photocopy a piece of paper and then you photocopy the photocopy. You lose some of the information," Papernot said. Not only that, but Papernot's research has also found it can further encode the mistakes, bias and unfairness that's already baked into the information ecosystem.

If real human-crafted sentences remain a critical AI data source, those who are stewards of the most sought-after troves — websites like Reddit and Wikipedia, as well as news and book publishers — have been forced to think hard about how they're being used.

"Maybe you don’t lop off the tops of every mountain," jokes Selena Deckelmann, chief product and technology officer at the Wikimedia Foundation, which runs Wikipedia. "It’s an interesting problem right now that we’re having natural resource conversations about human-created data. I shouldn’t laugh about it, but I do find it kind of amazing."

While some have sought to close off their data from AI training — often after it's already been taken without compensation — Wikipedia has placed few restrictions on how AI companies use its volunteer-written entries. Still, Deckelmann said she hopes there continue to be incentives for people to keep contributing, especially as a flood of cheap and automatically generated "garbage content" starts polluting the internet.

AI companies should be "concerned about how human-generated content continues to exist and continues to be accessible," she said.

From the perspective of AI developers, Epoch's study says paying millions of humans to generate the text that AI models will need "is unlikely to be an economical way" to drive better technical performance.

As OpenAI begins work on training the next generation of its GPT large language models, CEO Sam Altman told the audience at a United Nations event last month that the company has already experimented with "generating lots of synthetic data" for training.

"I think what you need is high-quality data. There is low-quality synthetic data. There’s low-quality human data," Altman said. But he also expressed reservations about relying too heavily on synthetic data over other technical methods to improve AI models.

"There’d be something very strange if the best way to train a model was to just generate, like, a quadrillion tokens of synthetic data and feed that back in," Altman said. "Somehow that seems inefficient."

Читайте на 123ru.net


Новости 24/7 DirectAdvert - доход для вашего сайта



Частные объявления в Вашем городе, в Вашем регионе и в России



Smi24.net — ежеминутные новости с ежедневным архивом. Только у нас — все главные новости дня без политической цензуры. "123 Новости" — абсолютно все точки зрения, трезвая аналитика, цивилизованные споры и обсуждения без взаимных обвинений и оскорблений. Помните, что не у всех точка зрения совпадает с Вашей. Уважайте мнение других, даже если Вы отстаиваете свой взгляд и свою позицию. Smi24.net — облегчённая версия старейшего обозревателя новостей 123ru.net. Мы не навязываем Вам своё видение, мы даём Вам срез событий дня без цензуры и без купюр. Новости, какие они есть —онлайн с поминутным архивом по всем городам и регионам России, Украины, Белоруссии и Абхазии. Smi24.net — живые новости в живом эфире! Быстрый поиск от Smi24.net — это не только возможность первым узнать, но и преимущество сообщить срочные новости мгновенно на любом языке мира и быть услышанным тут же. В любую минуту Вы можете добавить свою новость - здесь.




Новости от наших партнёров в Вашем городе

Ria.city

Запасы марганцевой руды в Запорожской области могут избавить Россию от импорта сырья

Семь заявок подали из Орловской области на Знание.Премия — 2024

Тропический лес на стене Московского зоопарка: в рамках фестиваля «Лето в Москве. Сады и цветы» реализуют необычные проекты

Президент Палестины Махмуд Аббас посетит Москву

Музыкальные новости

Обложка песни. Обложки альбомов песен. Сделать обложку для песни.

«Крылья Советов» и воронежский «Факел» сойдутся в битве аутсайдеров РПЛ

Депутат Госдумы Толмачев ответил Галкину* на критику Лепса и Газманова

Бойца ММА Исмаилова будут судить за унижение человеческого достоинства

Новости России

Президент Палестины Махмуд Аббас посетит Москву

Нгамале рассказал об оскорблениях после пенальти в матче РПЛ с «Зенитом»

В аэропорту Благовещенска 300 пассажиров более суток не могут улететь в Москву

Борьбу с «серыми перевозчиками» строительных отходов усилят в Подмосковье

Экология в России и мире

Я всегда с собой беру: аптечка в дорогу, составленная на основе неожиданных историй из отпуска

Мари Краймбрери, Звонкий, DAASHA выступят на марафоне «Авторадио»

Как записать дядю в телефоне

Большой стадионный концерт MACAN в Москве при поддержке Like FM

Спорт в России и мире

Теннисистка Самсонова вышла в четвертьфинал турнира WTA 1000 в Торонто

Калинская снялась с матча третьего круга турнира WTA 1000 в Торонто

Зверев: в Токио организация Олимпиады была намного лучше, чем в Париже

Теннисист Рублев обыграл Янника Синнера в 1/4 финала"Мастерса" в Монреале

Moscow.media

Квартальная выручка Freedom Holding Corp. выросла до $450,7 млн

Дешевле не будет. Цену непопулярного китайского кроссовера снизили на 27%

43664

Власти страны отказываются от легких водительских денег, VW скандалит, АВТОВАЗ торопится: итоги недели











Топ новостей на этот час

Rss.plus






Между Белгородом и Москвой назначили дополнительные поезда

Артист Грек пожертвовал миллионы на борьбу с искусственным интеллектом, чтобы защитить будущее

Нгамале рассказал об оскорблениях после пенальти в матче РПЛ с «Зенитом»

Реабилитация эвакуированных из Курской области детей началась в Подмосковье