{
  "id": 5142807,
  "title": "These Russian Mathematicians Taught AI Models How to Talk to Each Other Without Using Words",
  "url": "https://urgent.news/2026/09/02/these-russian-mathematicians-taught-ai-models-how-to-talk-to-each",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-02T18:20:00.000Z",
  "source": {
    "name": "Wired Business",
    "slug": "wired-business",
    "url": "https://www.wired.com/story/russian-startup-mostik-ai-models-communication/"
  },
  "original_language": "en",
  "account": "Russian mathematicians at startup Mostik have developed a groundbreaking technique that enables AI models to communicate with each other without using words. By leveraging the mathematical values within model weights, the company has created a bridge that allows larger models to transfer their knowledge to smaller models, significantly enhancing their intelligence. This innovative approach has propelled Mostik's model to the top of the challenging ARC-AGI 3 competition, which is notoriously difficult for AI systems to conquer.\n\nThe startup's technique allows a smaller model to benefit from the capabilities of a larger model, resulting in a more efficient and cost-effective solution. To illustrate the concept, Mostik created a bridge between two open-weight Chinese models: the 753-billion-parameter GLM-5.2 and a 4-billion-parameter version of Qwen-3.5. The resulting hybrid system costs one-twentieth of the full GLM model and performs at the average level of both, demonstrating the potential for cost savings and improved performance.\n\nMostik's CEO, Sasha Malysheva, explained that combining multiple AI models often yields better results, similar to how guessing the weight of a pig becomes more accurate when several people's guesses are averaged. This approach could increase the value of open-weight models, allowing them to compete more effectively against closed, proprietary models from leading AI labs.\n\nThe team's approach has already garnered attention from industry experts, such as Karl Tuyls from Google DeepMind and Stanislav Smirnov, a professor at the University of Geneva and a Fields Medalist. Smirnov acknowledges the difficulty in finding common ground between AI models, but believes that Mostik's technique could reveal new insights into AI model functionality and potentially compare AI reasoning to human cognition. Malysheva, the startup's CEO, persevered despite initial skepticism, proving that her talent for math was not underestimated.",
  "summary": "A startup called Mostik has a wild new approach to combining the capabilities of AI models.",
  "key_points": [
    "Russian mathematicians at startup Mostik developed AI communication technique.",
    "Technique enables knowledge transfer between AI models without words.",
    "Hybrid system costs 20% of full GLM model while matching average performance."
  ],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 2,
    "also_reported_by": [
      {
        "outlet": "Wired",
        "title": "These Russian Mathematicians Taught AI Models How to Talk to Each Other Without Using Words",
        "url": "https://urgent.news/2026/09/02/these-russian-mathematicians-taught-ai-models-how-to-talk-to-each-5143982",
        "published": "2026-09-02T18:20:00.000Z"
      }
    ]
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}