This Startup Wants to Spark a US DeepSeek Moment
From DeepSeek burst onto the scene in January as momentum grew around open source Chinese AI models. Some researchers are arguing for an even more open approach to AI creation, one that would spread model creation across the globe.
Prime Intellect, a startup specializing in decentralized AI, is currently training a large-scale language model, called INTELLECT-3, using a new type of distributed reinforcement learning to refine it. The model will demonstrate a new way to create open, competitive AI models using a range of hardware in different locations, without relying on large technology companies, said Vincent Weisser, the company’s CEO.
Weisser says the AI world is currently divided between those relying on closed US models and those using open Chinese offerings. The technology that Prime Intellect is developing democratizes AI by enabling more people to create and modify advanced AI themselves.
Improving AI models is no longer simply about increasing training data and calculations. 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? Make him improve by practicing in an environment where you can measure success and failure.
“These reinforcement learning environments are now a bottleneck for actually scaling capabilities,” Weisser tells me.
Prime Intellect has created a framework that allows anyone to create a personalized reinforcement learning environment for a particular task. The company combines the best environments created by its own team and the community to tune INTELLECT-3.
I tried running an environment for solving Wordle puzzles, created by Prime Intellect researcher Will Brown, by watching a small model solve Wordle puzzles (it was more methodical than me, to be honest). If I were an AI researcher trying to improve a model, I would spin up a bunch of GPUs and practice the model over and over while a reinforcement learning algorithm changed its weights, thereby turning the model into a Wordle master.
