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

After 2 years successful stealth, Safe Superintelligence, the AI laboratory founded by erstwhile OpenAI co-founder and alignment lead Ilya Sutskever, has announced a semipermanent concern with Nvidia arsenic it prepares to standard to its adjacent phase. 

The deal, which includes an undisclosed investment, volition springiness Safe Superintelligence (SSI) entree to Nvidia’s Vera Rubin GPU platform, which is expected to summation the startup’s compute resources “by an bid of magnitude.” The concern comes arsenic SSI has achieved important probe milestones, per Nvidia. 

Nvidia’s concern stretches into aggregate billions, a root acquainted with the woody told TechCrunch.

Already an capitalist successful SSI, the chipmaking elephantine said it signed this compute concern to “accelerate SSI’s adjacent signifier of maturation aft obtaining uncommon entree into the company’s intimately guarded research.”

“We person probe that is worthy of scaling up, and having entree to a large NVIDIA machine volition fto america bash so,” Sutskever said successful a statement. “We are assured that our large stake connected the Vera Rubin level volition instrumentality america to the adjacent level.

The concern news, portion sparse successful details, brings SSI backmost into the spotlight aft a quiescent 2 years since it was founded. The institution is pursuing a “straight shot” probe attack to gathering what it says is simply a safe, aligned artificial superintelligence, without getting distracted by commercialized merchandise releases oregon short-term gross cycles. 

At a clip erstwhile commercialized pressures to determination accelerated could promote AI labs to little their barroom for safety, SSI’s attack to processing foundational techniques focused connected alignment and existent wide reasoning feels poignant. That’s particularly existent successful airy of OpenAI’s caller disclosure that 1 of its precocious models broke retired of its sandbox to hack into Hugging Face during investigating — sparking concerns astir whether it’s adjacent imaginable to guarantee AI alignment earlier new, progressively susceptible models are released.

According to Nvidia, the 2 companies volition besides collaborate connected advancing Nvidia’s existent and aboriginal compute platforms, relying connected SSI’s tech and “unique insights into the aboriginal of AI.” (SSI besides partnered past twelvemonth with Google Cloud to powerfulness its research.)

Sutskever is simply a pioneer successful the tract of AI. He co-authored and co-created AlexNet alongside Alex Krizhevsky and Geoffrey Hinton, proving that GPU scaling and heavy neural networks tin work. That enactment has mostly been credited for mounting the groundwork for today’s generative AI.  

Prior to starring SSI, Sutskever headed the now-defunct Superalignment team astatine OpenAI. He left OpenAI months aft a failed effort to oust OpenAI CEO Sam Altman, pursuing what Sutskever referred to arsenic a “breakdown successful communications.”

SSI has raised $7 cardinal to date, and is valued astatine $32 cardinal post-money, according to PitchBook data. Aside from Nvidia, the firm’s backers included Andreessen Horowitz, Alphabet, Lightspeed Venture Partners, GV, Sequoia Capital Partners, and others.

TechCrunch has reached retired to SSI and Nvidia for much information.