🚀 AI start-up in the Chinese mainland, DeepSeek, just dropped R1 — the first major large language model (LLM) to get peer-reviewed. It’s setting a new benchmark in AI!
Released in January, R1 was built to ace reasoning-intensive tasks like math and coding. It’s a lean, cost-effective challenger to models from U.S. technology firms.
As an open-weight model, R1 can be downloaded for free. It’s already the most popular LLM on Hugging Face, with over 10.9 million downloads to date! 💾
Nature highlighted R1 as the first major LLM to undergo formal peer review. "This is a very welcome precedent," said Lewis Tunstall, a machine-learning engineer at Hugging Face. Sharing the process publicly helps us spot potential risks early.
DeepSeek also revealed R1’s price tag: about $294,000 for training — a fraction of the tens of millions often spent on similar models. Building the base model cost around $6 million.
The magic behind R1? A pure reinforcement learning approach. Instead of relying on human-selected reasoning examples, the model learned through automated trial and error. It even checks its own outputs using group relative policy optimization for extra efficiency.
Now, researchers worldwide are exploring these methods to boost the reasoning skills of other LLMs and expand into fields beyond math and coding. As Lewis Tunstall puts it, R1 has truly kick-started a revolution! 🔥
For students, entrepreneurs, and AI fans across Asia and beyond, R1 is a clear sign that powerful, transparent models can emerge from anywhere — and at a fraction of the cost. The AI game is changing fast!
Reference(s):
DeepSeek's R1 sets benchmark as first peer-reviewed major AI LLM
cgtn.com
