Ilya Sutskever Launches Safe AI Startup Backed by Nvidia with Imminent Model Release
Nearly three years ago, Ilya Sutskever, OpenAI's chief scientist and the mind behind ChatGPT, clashed with OpenAI co-founder Sam Altman over the organization's direction. Sutskever advocated for limiting language model development to ensure safety and prevent public risk, while Altman supported transforming OpenAI into a for-profit company. Ultimately, investors sided with Altman, reinstating him as CEO and approving the for-profit shift, leading Sutskever to leave and start a new AI venture.
In 2024, Sutskever co-founded Safe Superintelligence (SSI) with Daniel Levy and Daniel Gross, aiming to develop a language model focused on safety, reliability, and alignment with human values. SSI has raised $7 billion, reaching a valuation of $32 billion, and employs over 50 people across San Francisco and Tel Aviv, including several Israelis. Major investors include Nvidia, Alphabet, Sequoia, SV Angel, and Israeli funds like Overcrowd and Ventures Israel. Nvidia alone invested approximately $5 billion, likely through GPU purchases.
SSI's approach contrasts with competitors like Anthropic and OpenAI by emphasizing real-time learning and safety rather than merely increasing computational power or model size. According to insiders, SSI plans to launch its first language model imminently, possibly within the coming week. Some AI experts suggest the model is already in confidential testing. Dr. Hadami Nouri from the University of Toronto, where Sutskever earned his PhD, revealed he accessed an unnamed company's model that demonstrated unlimited context awareness, recalling 15 years of his writings.
SSI reportedly addresses the "knowledge freeze" problem in current language models, which cannot learn from new information after initial training. Their model can update its internal parameters dynamically while processing new data, potentially enhancing safety and efficiency. This breakthrough could reduce the need for massive computational resources, challenging the dominance of large-scale AI models.
However, some experts, like AI consultant Uri Aleibev, caution that SSI's continuous learning might introduce unpredictable behavior, and the model may initially serve niche applications rather than replace leading AI systems. The company remains tight-lipped, but investor pressure mounts for tangible progress. SSI's success could revolutionize AI development by balancing adaptability with safety and reducing hardware demands.
