Chinese AI firm DeepSeek has made breakthroughs in efficient, open-source AI frameworks. These advancements offer key lessons for Europe’s AI innovators.

As our eeNews Europe colleague Nick Flaherty reported, DeepSeek — which is headquartered in Hangzhou, China — has developed two AI frameworks capable of running large language models (LLMs) that rival those of OpenAI, Perplexity, and Google — using significantly fewer computing resources. The company employs unsupervised reinforcement learning to enhance the reasoning capabilities of its AI models, and has released its technology as open source under the MIT license, Flaherty noted.
 
AI news - DeepSeek

DeepSeek LLMs

DeepSeek's LLMs, which can handle up to 70 billion parameters, are optimized to run on Nvidia H100 GPUs, Flaherty explained. These GPUs, while powerful, are considered lower-performing compared to chips barred from export to China under U.S. government restrictions. Reports suggest that DeepSeek has access to as many as 50,000 H100 processors.

For those interested in the underlying technology, the groundbreaking paper on DeepSeek’s advancements is available online.

“DeepSeek is not the first to show that a talent-dense team can go toe-to-toe with the leading, most capitalized AI model companies," said Walter Goodwin, CEO and Founder of UK AI startup Fractile which recently saw investment from Pat Gelsinger, former CEO of Intel. "In Europe, Mistral was able for much of 2024 to provide open source models that rivaled Meta’s open Llama models, yet were trained on a fraction of the budget."

Subscribe
Tag alert: Subscribe to the tag Embedded & AI and you will receive an e-mail as soon as a new item about it is published on our website!

“Europe has a high talent density and is less constrained on compute availability than China, and so DeepSeek should be a wake-up call that proves Europe can also afford to play at the leading edge of AI.”

The open-source nature of DeepSeek’s frameworks has already impacted US-based competitors that monetize their AI chatbot services, Flaherty reported. In China, WiMi Hologram Cloud is developing intelligent programming tools powered by DeepSeek. These tools aim to assist programmers by completing code, analyzing quality, and suggesting optimizations, streamlining the development process and improving outcomes.

Popularity and Potential

DeepSeek’s popularity has surged over the past few days, with its chat app garnering 2.6 million downloads. However, sign-ups were paused following a reported cyberattack, Flaherty noted.

Nigel Toon, CEO of UK AI chip designer GraphCore, has also highlighted DeepSeek’s potential.

“DeepSeek AI’s breakthroughs, leveraging reinforcement learning and a diverse mixture-of-experts model, go beyond what has been achieved with single large models, all while being far more efficient,” Toon remarked. “While export restrictions on GPUs may have been a constraint, they have driven innovation, proving necessity is the mother of invention.”
Refer to eeNews Europe's article for more information. eeNews Europe is an Elektor International Media publication.
Subscribe
Tag alert: Subscribe to the tag Artificial Intelligence and you will receive an e-mail as soon as a new item about it is published on our website!