One of the co-founders of OpenAI, Ilya Sutskever, announced last Wednesday that he was going to start a new artificial intelligence company. This company would be called Safe Superintelligence Inc (SSI) and its goal would be to develop a “nuclear” security AI.
Also listed as co-founders are Daniel Levy, former engineer at OpenAI, and Daniel Gross, co-founder of Cue and former head of AI at Apple. In addition, the company will have offices in Palo Alto (California) and Tel Aviv (Israel). The announcement of the creation of SSI was made by Sutskever himself through a post published on X.
On their website, we can also find a statement where they explain that they have “launched the world’s first Superintelligence Lab (SSI), with a goal and a product: a secure superintelligence.” They claim that, according to their business model, “security, protection, and progress are isolated from short-term commercial pressures.”
What exactly makes one AI safer than another? At the moment, it is not very clear. In statements made to Bloomberg, Sutskever suggests that he will try to achieve safety through engineering advances incorporated into AI technology, rather than relying on on-the-fly fixes. To illustrate this, he makes a comparison with nuclear energy: “by safety, we mean a nuclear-like safety, not a ‘trust and safety’ type of safety.”
