Ilya Sutskever

Ilya Sutskever (born 1986) was centrally involved in three of the hinges on which modern deep learning turns: AlexNet in 2012, sequence-to-sequence learning in 2014, and the large language models of OpenAI. He was also one of the directors who voted to remove Sam Altman in November 2023, and he left the company six months later to start a laboratory of his own.

Background

Sutskever was born in Gorky in the Soviet Union — now Nizhny Novgorod, Russia — emigrated to Israel with his family at the age of five and grew up in Jerusalem. At sixteen he moved to Canada and entered the University of Toronto, where he took a bachelor's degree in mathematics (2005), a master's in computer science (2007), and a doctorate (2013). His doctoral advisor was Geoffrey Hinton.

AlexNet and seq2seq

In 2012, in Hinton's group, he and Alex Krizhevsky built the convolutional network that won ILSVRC 2012. That result moved computer vision from hand-engineered features to deep learning. DNNresearch, founded immediately afterwards, was acquired by Google in 2013, and Sutskever became a research scientist at Google Brain.

There, with Oriol Vinyals and Quoc V. Le, he published "Sequence to Sequence Learning with Neural Networks" (NIPS 2014), which mapped a variable-length input sequence to a fixed-dimensional vector with one LSTM and decoded the output sequence with another. NeurIPS gave the paper a Test of Time award in 2024, calling it "the cornerstone work that set the encoder-decoder architecture, inspiring later attention-based improvements leading to today's foundation model research".

OpenAI, 2015–2024

The founding announcement of 11 December 2015 named Sutskever OpenAI's research director; he is also one of its listed authors. He later became chief scientist, directing the research behind the GPT series.

On 5 July 2023 OpenAI announced the Superalignment team, co-led by Sutskever and Jan Leike. The stated goal was the scientific and technical breakthroughs needed "to steer and control AI systems much smarter than us", to be solved within four years, backed by 20 percent of the compute the company had secured to that date. The line connecting his research position to what he did next runs through that announcement.

November 2023

When the removal was announced, the OpenAI board consisted of Sutskever as chief scientist together with the independent directors Adam D'Angelo, Tasha McCauley, and Helen Toner; Altman left the board as well, and Greg Brockman stepped down as its chairman. The five days that followed are covered in The OpenAI Coup. On 20 November Sutskever posted: "I deeply regret my participation in the board's actions. I never intended to harm OpenAI." He also signed the employee letter demanding the board resign. He left the board as part of the settlement that returned Altman.

The longest account by a participant is the roughly ten-hour deposition he gave on 1 October 2025 in Musk v. Altman, whose 365-page transcript was released that November. Decrypt, reporting from the transcript, wrote that Sutskever testified to having sent the independent directors a 52-page document criticising Altman's leadership. It is sworn testimony from one side, not an agreed account of the episode.

Leaving, and Safe Superintelligence

On 14 May 2024 OpenAI announced Sutskever's departure and the appointment of Jakub Pachocki as chief scientist. Altman's note to the company called him "easily one of the greatest minds of our generation, a guiding light of our field, and a dear friend".

On 19 June 2024 Sutskever founded Safe Superintelligence Inc. (SSI) with Daniel Gross and Daniel Levy. Its founding statement describes a laboratory with "one goal and one product: a safe superintelligence", carrying no product cycles or management overhead so that safety and progress stay insulated from short-term commercial pressure; its offices are in Palo Alto and Tel Aviv. On 4 September 2024 it announced US$1 billion raised from NFDG, a16z, Sequoia, DST Global, and SV Angel. On 3 July 2025, after Gross left, Sutskever became formally chief executive with Levy as president.

On 27 July 2026 SSI and NVIDIA announced a long-term strategic partnership giving SSI access to NVIDIA's Vera Rubin platform and raising its compute by an order of magnitude. NVIDIA also invested; neither company disclosed a figure, though TechCrunch reported it ran into multiple billions and noted Bloomberg's figure of US$5 billion. Sutskever's statement was characteristically narrow: "We have research that is worthy of scaling up, and having access to a big NVIDIA computer will let us do so." Two years after founding, a lab that has never shipped a product now has a chipmaker behind it.

Awards

• 2024 NeurIPS Test of Time Award (seq2seq, with Vinyals and Le)
• 2026 NAS Award for the Industrial Application of Science, given that year in artificial intelligence

Legacy

At the NeurIPS award talk in Vancouver on 13 December 2024, Sutskever said that "pre-training as we know it will unquestionably end" and that "we've achieved peak data and there'll be no more" — the internet is finite, and training data behaves like a fossil fuel. Next-generation systems, he predicted, would be genuinely agentic and would reason — and the more they reasoned, the less predictable they would become. He belongs to Hinton's line, but takes a sharper position within it: the technical problem worth working on is not raising capability but building it in a form that stays controllable. Whether SSI can demonstrate that as an artefact rather than an argument is, as of 2026, still unknown.