Добавить новость
ru24.net
News in English
Декабрь
2024
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
18
19
20
21
22
23
24
25
26
27
28
29
30
31

Tech Giants Unveil New AI Training Techniques To Overcome Data Scarcity Constraints

0
eWeek 

Artificial intelligence (AI) is hitting a turning point as companies like OpenAI tackle significant challenges like skyrocketing energy use and spiraling costs by focusing on smarter, more efficient systems that think a bit more like humans. Among the issues they’re facing is a limitation of training data, with most of what’s online already having been used.

Take OpenAI’s “o1” model, for example—instead of throwing more computing power at the problem, it approaches tasks like humans might, breaking them into smaller steps and learning from expert feedback. A method called “test-time compute” lets it concentrate resources where they’re needed most, which bolsters performance without a massive price tag.

At a recent TED AI event, OpenAI researcher Noam Brown showed how a poker bot used this approach, reasoning for just 20 seconds to achieve the same results as a model trained for 100,000 times longer.

Addressing the Sustainability Challenge

This is a game-changer for AI development. Training today’s massive language models (LLMs) eats up enormous amounts of energy, straining power grids and raising environmental red flags. On top of that, the well of high-quality data is running dry. The o1 model is designed to handle these hurdles by relying on carefully chosen data and using resources more thoughtfully.

OpenAI isn’t alone. Other big players like Google DeepMind, Anthropic, and xAI are working on similar techniques. This competition is, of course, giving the whole race rocket boosters. The upside? These advances could lower costs and make cutting-edge AI tools more accessible to industries that haven’t been able to afford them.

This shift is also shaking up the hardware market. Companies like Nvidia, which dominates AI chip production, might need to rethink their game as these new, more efficient methods gain traction. With the door open to innovation, new competitors could start carving out a space in the market.

Looking Ahead

The big takeaway? AI’s focus has shifted. It’s less about building bigger models and more about building better ones—systems that reason, adapt, and learn smarter, not harder. As OpenAI co-founder Ilya Sutskever put it, “Scaling in the right direction is what matters most.”

We’re officially entering a new era for AI—one that promises smarter, more sustainable systems and a more competitive landscape. The possibilities are huge, and this is just the beginning.

The post Tech Giants Unveil New AI Training Techniques To Overcome Data Scarcity Constraints appeared first on eWEEK.




Moscow.media
Частные объявления сегодня





Rss.plus




Спорт в России и мире

Новости спорта


Новости тенниса
ATP

Новак Джокович заявился на турнир ATP-500 в Дохе






Telegram-каналы сообщают, что в результате взрыва в Москве погиб генерал-лейтенант Кириллов

Начальник войск РХБЗ Игорь Кириллов погиб при взрыве в Москве

Два человека погибли после взрыва автомобиля на Рязанском проспекте в Москве

Ван И: отношения Китая и России не направлены против третьих сторон