Ilya Sutskever

wiki
, , , ,
1

Israeli-Canadian deep learning researcher; co-inventor of AlexNet, co-founder and longtime Chief Scientist of OpenAI, and founder and CEO of Safe Superintelligence Inc.

Basic Information / Profile

Field Details
Full Name Ilya Efimovich Sutskever
Born 1986, Gorky (now Nizhny Novgorod), Russian SFSR, Soviet Union
Nationality Israeli-Canadian
Current Institution Safe Superintelligence Inc. (SSI)
Current Title CEO
Research Fields Deep learning, recurrent neural networks, sequence modelling, AI alignment, superintelligence safety
PhD Advisor Geoffrey Hinton
PhD Thesis Training Recurrent Neural Networks (University of Toronto, 2013)
Personal Website cs.toronto.edu/~ilya
X / Twitter @ilyasut
Google Scholar scholar.google.com/citations?user=x04W_mMAAAAJ

Overview

Ilya Sutskever is an Israeli-Canadian computer scientist widely regarded as one of the most consequential figures in the deep learning era. A student of Geoffrey Hinton at the University of Toronto, he co-invented AlexNet — the convolutional network whose ImageNet victory in 2012 is commonly cited as the inflection point of modern AI — and later co-authored the sequence-to-sequence framework that underpins much of contemporary natural language processing. As co-founder and Chief Scientist of OpenAI from 2015 to 2024, he shaped the research agenda that produced GPT, ChatGPT, DALL-E, CLIP, and the o1 reasoning model, and is credited with establishing the organisation’s scaling ethos. In June 2024 he founded Safe Superintelligence Inc. (SSI), a company whose singular stated objective is to build safe superintelligence, and became its CEO in July 2025 after co-founder Daniel Gross departed for Meta. He is one of the most-cited computer scientists in history and has won the NeurIPS Test of Time Award three consecutive times (2022–2024).

Early Life & Education

Childhood: Russia, Israel, Canada

Sutskever was born in 1986 into a Jewish family in Gorky (now Nizhny Novgorod), Russian SFSR, Soviet Union. At the age of five, his family made aliyah and settled in Jerusalem, where he attended school and began studies at the Open University of Israel. At around sixteen the family relocated again, this time to Canada. He attended a Canadian high school for approximately one month before being admitted to the University of Toronto directly as a third-year undergraduate — an admission that reflects the advanced level he had already reached.

University of Toronto — BSc, MSc, PhD (2002–2013)

Sutskever completed three degrees at the University of Toronto under the intellectual influence of Geoffrey Hinton, one of the central architects of modern deep learning. He received a Bachelor of Science in Mathematics in 2005, a Master of Science in Computer Science in 2007 (thesis: Nonlinear Multilayered Sequence Models), and a PhD in Computer Science in 2013 (thesis: Training Recurrent Neural Networks). Across this decade he contributed to Hinton’s research group on restricted Boltzmann machines, deep belief networks, and the early foundations of large-scale neural network training. In 2012, during the final phase of his doctorate, he co-built AlexNet with Hinton and Alex Krizhevsky, winning the ImageNet Large Scale Visual Recognition Challenge by a margin that shocked the computer vision community.

Postdoctoral Interlude and DNNResearch (2012–2013)

Following the AlexNet result, Sutskever spent approximately two months as a postdoctoral researcher with Andrew Ng at Stanford before returning to Toronto to join DNNResearch, a spinoff of Hinton’s group. In March 2013, Google acquired DNNResearch and hired Sutskever, Hinton, and Krizhevsky — an acquisition widely described as the opening move of the deep learning talent wars.

Career

Google Brain — Research Scientist (2013–2015)

At Google Brain, Sutskever worked with Oriol Vinyals and Quoc V. Le to develop the sequence-to-sequence (seq2seq) learning framework, presented at NeurIPS 2014. The architecture — training one recurrent neural network to encode a variable-length input sequence and a second to decode it into an output sequence — provided the foundation for neural machine translation, dialogue systems, and later language modelling at scale. He also contributed to TensorFlow and appeared as a co-author on the AlphaGo paper. He left Google at the end of 2015 to co-found OpenAI.

OpenAI — Co-founder and Chief Scientist (2015–2024)

Sutskever was among the founding group that launched OpenAI in December 2015, alongside Sam Altman, Greg Brockman, Elon Musk, and others. As Chief Scientist he had direct oversight of OpenAI’s research direction for nearly nine years, a tenure that spanned the field’s most consequential period. He established the organisation’s scaling ethos — the conviction that increasing model size and data volume would yield proportionally improved capabilities — which guided the GPT series from GPT-1 through GPT-4. He was a key figure in the research leading to ChatGPT and later led the team that produced the o1 reasoning model, which used reinforcement learning to elicit chain-of-thought problem-solving at test time. In July 2023, he co-announced the Superalignment project, committing to solve the alignment of superintelligent systems within four years and pledging 20% of OpenAI’s compute to the effort — a commitment that, according to subsequent reporting, was never fully delivered.

In November 2023, Sutskever was one of four board members who voted to remove Sam Altman as CEO, citing concerns about candour and governance. He subsequently authored a 52-page memo to the board relying on information provided by Mira Murati that detailed his objections to Altman’s conduct. Within days he publicly expressed regret at having participated in the firing; Altman was reinstated after a week, and Sutskever stepped down from the board. He was largely absent from OpenAI operations in the months that followed. In May 2024 he announced his departure, describing his next project as “very personally meaningful.” His resignation came on the same day as that of Jan Leike, co-lead of the Superalignment team, who cited an erosion of safety culture at OpenAI.

Safe Superintelligence Inc. — Co-founder and CEO (2024–present)

In June 2024, Sutskever announced Safe Superintelligence Inc. (SSI), co-founded with Daniel Gross and Daniel Levy, with offices in Palo Alto and Tel Aviv. The company’s stated mission is narrow and explicit: its first and only product will be safe superintelligence; it will build no other products or pursue revenue in the interim. In September 2024, SSI raised $1 billion from Andreessen Horowitz, Sequoia Capital, DST Global, and SV Angel. By March 2025 a further $2 billion raise had pushed the reported valuation to $32 billion. In June 2025, Meta attempted to acquire SSI outright; the company declined. Shortly thereafter, co-founder and CEO Daniel Gross left for Meta, and Sutskever assumed the CEO role. The company has released no public research as of mid-2026 and operates with a high degree of secrecy.

Key Contributions

  • AlexNet (NeurIPS 2012, with Alex Krizhevsky and Geoffrey Hinton) — A deep convolutional neural network that won the 2012 ImageNet challenge with a top-5 error rate of 15.3%, nearly halving the previous year’s best result; widely considered the paper that catalysed the modern deep learning era. The source code was preserved by the Computer History Museum.
  • Dropout (JMLR 2014, with Hinton, Krizhevsky, Srivastava, and Salakhutdinov) — Co-authored the paper formalising dropout as a regularisation method for neural networks; became one of the most universally applied techniques in deep learning.
  • Sequence-to-Sequence Learning with Neural Networks (NeurIPS 2014, with Oriol Vinyals and Quoc V. Le) — Introduced the encoder-decoder recurrent architecture that formed the basis of neural machine translation and later shaped transformer-era language modelling. NeurIPS Test of Time Award 2022.
  • GPT series and ChatGPT — As Chief Scientist of OpenAI, provided research leadership across the GPT-1 through GPT-4 lineage and the deployment of ChatGPT; credited with institutionalising the scaling hypothesis as the organisation’s central research commitment.
  • DALL-E and CLIP — Co-author on the foundational vision-language papers that established contrastive and generative approaches to multimodal AI; NeurIPS Test of Time Award 2023 (for related work) and 2024.
  • o1 reasoning model — Led the OpenAI research effort producing the o1 model, which applies reinforcement learning to elicit chain-of-thought reasoning at inference time, representing a distinct paradigm shift from pure next-token prediction.
  • Superalignment — Co-announced and helped conceptualise OpenAI’s programme to align superintelligent systems, bringing institutional focus to the technical alignment problem at a time of rapid capability scaling.
  • Safe Superintelligence Inc. — Founded SSI as a dedicated safety-first AI lab with no commercial product obligations, a structural bet that separating capability research from revenue pressure is a prerequisite for safe development.

Awards & Recognition

  • MIT Technology Review Innovators Under 35 (2015) — Recognised for contributions to deep learning and neural network research.
  • Fellow of the Royal Society (FRS) (2022) — Elected to the UK’s national academy of sciences, one of the oldest and most prestigious scientific fellowships.
  • TIME100 AI (2023, 2024) — Named to Time magazine’s annual list of the most influential people in artificial intelligence in consecutive years.
  • NeurIPS Test of Time Award (2022, 2023, 2024) — Won in three consecutive years, recognising papers with lasting impact on the field over at least a decade; the only researcher to achieve this distinction three times in a row.
  • Honorary Doctorate, University of Toronto (2025) — Awarded by his alma mater for leadership in AI and responsible AI development.
  • National Academy of Sciences Award for the Industrial Application of Science (2026) — The first time this award was presented in the field of artificial intelligence.

Key Relationships

  • Geoffrey Hinton — Doctoral advisor and the most direct intellectual influence on Sutskever’s research formation; co-inventor of AlexNet; after winning the 2024 Nobel Prize in Physics, Hinton publicly stated he was proud that Sutskever had participated in the vote to remove Sam Altman, citing shared AI safety concerns.
  • Alex Krizhevsky — University of Toronto labmate and co-inventor of AlexNet; the three-way collaboration between Sutskever, Krizhevsky, and Hinton produced the paper that began the modern deep learning era.
  • Sam Altman — Co-founder and CEO of OpenAI; Sutskever’s close collaborator for nearly nine years before the November 2023 board crisis, in which Sutskever voted for Altman’s removal and later expressed public regret.
  • Oriol Vinyals — Co-author on the sequence-to-sequence paper at Google Brain; later led AlphaStar and other major DeepMind projects.
  • Quoc V. Le — Co-author on the seq2seq paper; long-time Google Brain researcher and collaborator on early large-scale deep learning work.
  • Jan Leike — Co-lead of OpenAI’s Superalignment project; his simultaneous departure with Sutskever in May 2024 and his public statement citing erosion of safety culture framed the exits as a joint signal about OpenAI’s direction.
  • Daniel Gross — Co-founder and original CEO of SSI; departed for Meta in June 2025 following Meta’s unsuccessful acquisition bid, leaving Sutskever as sole CEO.
  • Daniel Levy — Co-founder of SSI; researcher and former OpenAI colleague who joined Sutskever to build the technical team.

Personal Style

Sutskever is known for combining deep mathematical intuition with an almost mystical commitment to the long-run safety question — a combination that makes him unusual among researchers equally at home in benchmark-driven engineering and existential risk reasoning. His public remarks are notably sparse and carefully hedged; he rarely gives interviews, and when he does, his statements tend toward the philosophical rather than the operational, as in his 2022 observation that large neural networks may be “slightly conscious,” which he offered not as a claim but as a possibility worth taking seriously. Within research communities he is described as having an unusually strong prior on what will scale — a calibration honed through the AlexNet experiment and subsequently the GPT series — which made him an authoritative internal voice on which research directions OpenAI should pursue. The founding structure of SSI, with no products, no revenue, and no public communications about research progress, is itself an expression of his style: maximalist in ambition, minimalist in public presence.

References