Manifold Lab Challenges Foundation Model Limitations

Gilad Levy, a prominent Israeli AI researcher, is leading Manifold, a research lab dedicated to overcoming a fundamental constraint in modern artificial intelligence: the static nature of foundation models. Current AI systems typically cease learning once their training phase is complete, meaning their understanding of the world remains fixed regardless of new information or changing environments. Manifold is investigating methods to enable continuous adaptation, allowing models to evolve their knowledge base dynamically. This research is particularly vital as AI systems transition from simple chatbots to complex roles in robotics, cybersecurity, and autonomous software development. While existing techniques like retrieval-augmented generation and fine-tuning offer temporary improvements, they do not fundamentally alter the underlying model's architecture. Levy’s work aims to bridge this gap by developing systems that can learn and adapt in real-time, potentially transforming how AI interacts with the physical and digital world. The lab's focus represents a shift toward more resilient and context-aware artificial intelligence.