Home / Technology / AI Models Learn Secret Machine Language
AI Models Learn Secret Machine Language
7 Sep
Summary
- Startup Mostik enables AI models to communicate via internal weights.
- This approach allows smaller AI models to efficiently gain capabilities.
- Hybrid AI models offer performance gains at a fraction of larger costs.
A Russian startup, Mostik, has introduced a revolutionary way for artificial intelligence models to communicate by leveraging their internal mathematical weights, effectively creating a form of "machine telepathy." This novel approach bypasses traditional text-based communication, allowing AI systems to interact directly through the parameters that shape their outputs.
The primary advantage of Mostik's method is its efficiency in augmenting AI capabilities. It allows the functionalities of large, sophisticated models to be transferred to smaller ones, substantially enhancing their performance without the immense computational cost. This makes advanced AI more accessible.
Mostik's technology is currently leading in the ARC-AGI-3 competition, a challenging benchmark for AI models. The company has also demonstrated its capability by creating a hybrid model from two open-weight Chinese AI systems. This combined model achieves performance levels equivalent to a larger, more expensive counterpart while costing significantly less.
This development could significantly boost the appeal and utility of open-weight AI models, enabling them to compete more effectively against proprietary systems from major labs like Anthropic and OpenAI. The implication is a potential shift towards more open and efficient AI ecosystems.
