Artificial intelligence has once again seen a major breakthrough. After leaving OpenAI as the chief scientist, Ilya Sutskever founded SSI (Safe Superintelligence), which has now released some details about its first model. According to recent revelations from multiple netizens on overseas social platforms, the institution is currently exploring a small reasoning engine based on TTT (Training While Testing), marking a significant step forward in its path toward safe superintelligence.

According to the information, the core of this new architecture lies in teaching the model "how to learn." Unlike traditional large models, where parameters remain unchanged during inference, this engine can dynamically update certain model weights while solving real problems. This design highly aligns with Ilya's core views, that true superintelligence should not rely solely on static training, but rather have the ability to continuously learn and evolve after deployment. Although TTT technology has long faced significant security challenges, it is reported that relevant technical difficulties have been resolved.

In terms of release schedule, this version of the model may be available for testing by a select group of invited users as early as August this year. Additionally, the next generation of the model is expected to scale up by 10 times. Although SSI has remained highly secretive so far, the content revealed this time closely matches previous industry clues regarding its technical roadmap, technical progress, and collaborations with giants like NVIDIA. With this new concept taking shape, the artificial intelligence industry may soon see a disruptive product capable of "changing tactics on the spot."