The Silent Conversation: How Russian Mathematicians Are Redefining AI Collaboration
There’s something profoundly intriguing about the idea of machines communicating without words. It’s like watching two minds connect on a level we can’t quite grasp—a kind of telepathy, but for algorithms. Recently, a team of Russian mathematicians at a startup called Mostik (Russian for bridge) has achieved just that. They’ve developed a method for AI models to interact using the mathematical values in their weights, bypassing the need for text-based output. Personally, I think this is a game-changer, not just for AI efficiency but for how we conceptualize machine intelligence itself.
What makes this particularly fascinating is the way it challenges our assumptions about AI scalability. For years, the industry has been obsessed with building larger, more data-hungry models. But Mostik’s approach suggests a different path: what if the future of AI isn’t about size, but about collaboration? Sasha Malysheva, Mostik’s CEO, believes this could be the case. She argues that combining models—rather than scaling them—might be the key to advancing AI. From my perspective, this is a bold claim, but it’s one that aligns with a broader trend in machine learning: the power of ensembles.
One thing that immediately stands out is the analogy Malysheva uses: guessing the weight of a pig. In math circles, it’s a classic example of how collective estimates often outperform individual expertise. This idea of collective intelligence is at the heart of Mostik’s innovation. By enabling models to “talk” without producing text, they’ve streamlined the process of combining outputs. What this really suggests is that AI collaboration doesn’t need to be clunky or resource-intensive. It can be elegant, efficient, and—dare I say—almost intuitive.
But let’s take a step back and think about the implications. If Mostik’s method takes off, it could democratize access to high-quality AI. Smaller, open-weight models could compete with the proprietary giants from labs like OpenAI or Anthropic. This raises a deeper question: could this shift the balance of power in the AI industry? I believe it could. By making collaboration more accessible, Mostik isn’t just bridging models—it’s bridging the gap between frontier labs and the rest of the world.
A detail that I find especially interesting is the hybrid system Mostik created between two Chinese models: GLM-5.2 and Qwen-3.5. The resulting system costs one-twentieth of the full GLM model but performs halfway between the two. This isn’t just a technical achievement; it’s a proof of concept for a new paradigm in AI development. What many people don’t realize is that this kind of efficiency could revolutionize industries that rely on AI, from healthcare to finance.
Vladimir Arustamian, tech lead at Lovable, points out that Mostik’s technique could enable the training of more specialized models. Imagine pairing a general AI with a domain-specific model in biology or physics. The possibilities are staggering. Karl Tuyls, a former Google DeepMind scientist, calls it a “no-brainer” for efficient model deployment. I couldn’t agree more. This isn’t just about saving time or money—it’s about unlocking new capabilities.
But here’s where it gets really intriguing: Mostik’s work might also shed light on how AI models think. Stanislav Smirnov, a Fields Medalist and Mostik’s chief scientist, suggests that their approach could reveal commonalities between AI reasoning and human cognition. If you take a step back and think about it, this could be the first step toward understanding the mind of a machine. What does it mean for an AI to “reason”? How does it compare to our own thought processes? These are questions that keep me up at night.
What’s equally inspiring is the story behind Mostik itself. Malysheva’s journey from being told she couldn’t solve Math Olympiad problems to leading this groundbreaking work is a testament to perseverance. When peers warned her the bridge approach was too difficult, she saw it as a challenge. In my opinion, this kind of determination is what drives innovation. It’s not just about solving problems—it’s about proving that the impossible is possible.
As I reflect on Mostik’s achievements, I can’t help but wonder: are we on the cusp of a new era in AI? One where collaboration, not competition, drives progress? Personally, I think we are. The silent conversation between AI models isn’t just a technical feat—it’s a metaphor for the future of intelligence itself. And if Mostik has anything to say about it, that future is going to be a lot more connected than we ever imagined.
