The first model of SSI founded by former OpenAI chief scientist Ilya was exposed, and the field of artificial intelligence once again ushered in core progress.

The first model of SSI founded by former OpenAI chief scientist Ilya was exposed, and the field of artificial intelligence once again ushered in core progress. SSI (Safe Superintelligence), founded by former OpenAI chief scientist Ilya Sutskever after his resignation, has received relevant news about its first model. According to the latest revelations from many netizens on overseas social platforms, the agency is currently actively exploring a small inference engine based on TTT (test-time training), which marks a substantial step towards safe superintelligence. It is understood that the core of this new architecture is to let the model learn "how to learn." Unlike traditional large models where the parameters remain unchanged during the inference process, this engine can dynamically update some model weights during the actual problem solving process. This design is highly consistent with Ilya’s previous core point of view, that is, true superintelligence must not rely solely on static training, but should have the ability to continue learning and evolving after deployment. Although TTT technology has always faced considerable challenges in terms of security, it was revealed that the relevant technical problems have been overcome. In terms of release pace, this version of the model is expected to be open for testing to a small number of specially invited users as soon as August this year. Additionally, its next-generation version is expected to achieve 10x scaling in model size. Although SSI has remained highly mysterious before, the content of this exposure is highly consistent with previous clues in the industry about its technical route, technological progress, and cooperation with giants such as Nvidia. With the implementation of this new concept, the artificial intelligence industry may usher in disruptive products that have the ability to "change their tactics when taking action". via AI News (author: AI Base)