Ilya Sutskever

Ilya Sutskever is an AI researcher, co-founder and Chief Scientist of Safe Superintelligence Inc. (SSI) since 2024, and co-founder and former Chief Scientist of OpenAI (2015–2024). He is a co-author of several foundational deep-learning papers, including AlexNet (2012), the Scaling Laws for Neural Language Models paper (2020), and the InstructGPT paper (2022), and he was among the OpenAI board members involved in the November 2023 firing of Sam Altman.

Background

Sutskever completed his PhD under Geoffrey Hinton at the University of Toronto and was a co-author on AlexNet (2012), the Krizhevsky/Sutskever/Hinton ImageNet paper that catalyzed the deep-learning era. He joined Google Brain in 2013 and co-founded OpenAI in 2015, serving as Chief Scientist until 2024. In November 2023 he was among the OpenAI board members who voted to fire Altman; after the reversal he publicly recanted, and he left OpenAI in May 2024. In June 2024 he co-founded SSI with Daniel Gross and Daniel Levy, a lab structured around a single product ("safe superintelligence") with no intermediate commercial releases.

Research contributions

Sutskever's research spans the foundations of modern deep learning. AlexNet (2012, with Krizhevsky and Hinton) was the ImageNet result that demonstrated the effectiveness of deep convolutional networks, reporting top-1 and top-5 error rates of 37.5% and 17.0% on LSVRC-2010 and a winning 15.3% ILSVRC-2012 top-5 error against the runner-up's 26.2% (ImageNet Classification with Deep Convolutional Neural Networks (Krizhevsky, Sutskever and Hinton, NeurIPS 2012)). His work on sequence-to-sequence learning (2014) became foundational to modern natural-language processing. He is a co-author of Scaling Laws for Neural Language Models (2020, Kaplan et al.), the empirical scaling-laws work on which scaling laws as a concept rests, and of the InstructGPT paper (Ouyang et al., 2022), which introduced RLHF as the dominant post-training paradigm.

At OpenAI he co-led the GPT-2, GPT-3, and GPT-4 research efforts. He led OpenAI's Superalignment team, launched in July 2023 and dissolved in mid-2024 following his and Jan Leike's departures.

Positions on AI safety and policy

Sutskever publicly committed to solving alignment for superintelligent systems; the Superalignment team was associated with OpenAI's pledge to dedicate 20% of its compute to the effort. His public statements after leaving OpenAI have been minimal, contrasting with the more vocal exit of former Superalignment colleague Leopold Aschenbrenner, whose Situational Awareness was dedicated to Sutskever.

SSI's stated thesis is to build a "straight-shot" safe superintelligence with no commercial intermediate products and no deployment-driven timeline pressure, which Sutskever and the company frame as a contrast with frontier labs whose safety work competes with product timelines.

In his research approach, Sutskever has historically emphasized empirical-scaling intuition over theoretical framings. In interviews he has cited consciousness, self-awareness, and generalization as open scientific questions, and has avoided the existential-risk rhetorical style associated with Hinton or Bengio while treating the underlying problems as serious.

Safe Superintelligence Inc.

Sutskever co-founded SSI in June 2024 with Daniel Gross and Daniel Levy. The company reportedly raised more than $1 billion at valuations in the tens of billions of dollars by 2025 despite having no product.

Relationships

• related: OpenAI — co-founder, former Chief Scientist
• related: Sam Altman — OpenAI co-founder; role in November 2023 crisis
• related: Scaling Laws Paper — co-author
• related: InstructGPT Paper — co-author
• related: Scaling Laws, RLHF — concepts his work founded
• related: Leopold Aschenbrenner — former Superalignment colleague; Situational Awareness was dedicated to Sutskever
• related: Jan Leike — Superalignment co-lead
• supports: AI Safety Frameworks, ambitious alignment research programs
• related: AGI Timelines