"Prime Intellect, a startup specializing in decentralized AI, is currently training a frontier large language model, called INTELLECT-3, using a new kind of distributed reinforcement learning for fine-tuning. The model will demonstrate a new way to build competitive open AI models using a range of hardware in different locations in a way that does not rely on big tech companies, says Vincent Weisser, the company's CEO."
"Improving AI models is no longer a matter of just ramping up training data and compute. Today's frontier models use reinforcement learning to improve after the pre-training process is complete. Want your model to excel at math, answer legal questions, or play Sudoku? Have it improve itself by practicing in an environment where you can measure success and failure. "These reinforcement learning environments are now the bottleneck to really scaling capabilities," Weisser tells me."
Momentum around open-source Chinese AI models has spurred interest in more distributed approaches to building AI. Prime Intellect is training a frontier large language model, INTELLECT-3, using distributed reinforcement learning for fine-tuning across diverse hardware and locations. The approach enables competitive open models without dependence on major tech companies. Reinforcement learning is used to improve models after pre-training by practicing in environments that measure success and failure. Reinforcement learning environments are identified as the bottleneck for scaling capabilities. A framework allows anyone to create customized reinforcement learning environments, and combined community and internal environments are being used to tune INTELLECT-3. A Wordle-solving environment demonstrated small models improving task performance through practice.
Read at WIRED
Unable to calculate read time
Collection
[
|
...
]