Ilya Delivers, Huang Jen-Hsun Bets 50 Billion on SSI, First Model Suspected to Launch This Week
Ilya Sutskever's Safe Superintelligence (SSI) lab, backed by a rumored $5 billion investment from NVIDIA granting exclusive access to next-gen systems, may release its first groundbreaking model as early as this week.
Industry insiders are hinting at a major breakthrough from a non-major player, with some claiming AI will be "utterly transformed." The model is speculated to fundamentally differ from current large language models like ChatGPT. Instead of relying on massive, static pre-training and ever-larger context windows, SSI's approach reportedly centers on **Test-Time Training (TTT)**. This would allow the model to **update its own internal weights in real-time** as it processes information, effectively "learning on the fly" and internalizing new knowledge like a human, rather than just referencing it from a large prompt.
This aligns with Ilya Sutskever's long-stated vision of moving beyond the "pre-training" and "scaling" era. He envisions superintelligence not as a monolithic, pre-loaded database, but as an adaptive, "infinitely curious 15-year-old genius" capable of continual learning through interaction. SSI's strategy has been focused solely on this goal of "safe superintelligence" through efficient continual learning and alignment, explaining its prior silence.
If successful, this model could disrupt the entire AI industry, challenging the current reliance on vast compute resources for training and the business models built around context windows. It suggests that core technological leaps, not just financial scale, remain decisive in the AI race.
marsbit6 хв тому