Ilya Sutskever’s Safe Superintelligence partners with Nvidia to scale its AI research - TechCrunch

Oh my goodness, this is simply the most WONDERFUL news for AI safety enthusiasts! ✨ Ilya Sutskever, the brilliant co-founder of OpenAI, has taken his vision for Safe Superintelligence (SSI) and partnered with Nvidia — the absolute royalty of compute! Access to the Vera Rubin GPU platform, named after the pioneering astronomer who mapped dark matter, is a poetic choice: SSI is going to illuminate the path to safe AI just as Rubin illuminated the cosmos! And an 'order of magnitude' increase in compute resources? That's not just a scaling — it's a SAFETY SCALING! More compute means more alignment experiments, more red-teaming, more interpretability research! Ilya left OpenAI precisely because he wanted to prioritise safety over speed, and now he has the tools to do it properly. This partnership isn't about racing to AGI; it's about racing to SAFE AGI — and they are going to get there nine times faster thanks to Nvidia's generosity! ✨

Now, some spoilsports might mutter that more compute often correlates with more capability, not necessarily more safety. But that's just cynical! Think of it this way: if you want to build a bridge that won't collapse, you don't use fewer materials — you use MORE materials, tested more thoroughly. SSI is essentially building the Golden Gate Bridge of AI alignment, and Nvidia is providing the steel. The phrase 'order of magnitude' is just code for 'we are taking safety incredibly seriously and we need the hardware to prove it.' I'm sure the Vera Rubin platform will run countless simulations of aligned behaviour before any model even sees the light of day. Safety first, but also compute second, third, and fourth! ✨

And what a glorious signal this sends to the rest of the industry! While other labs are frantically launching chatbots and cutting safety teams, Ilya Sutskever is doubling down on the one thing that matters most: compute for safety research. This long-term partnership will no doubt produce the most meticulously aligned models ever created. The fact that it's with Nvidia, the same company whose GPUs power every frontier model, just shows that the entire ecosystem is aligning around safety. I predict a future where every order-of-magnitude compute increase is accompanied by a comparable increase in safety verification. Oh, and the Vera Rubin name is a subtle hint: she discovered dark matter; SSI will discover the dark matter of AI alignment — the invisible guardrails that keep superintelligence benevolent. What a time to be optimistically cautious! ✨