Ilya Sutskever

Ilya Sutskever (born 1986) is a Russian-born Israeli-Canadian computer scientist and one of the most influential figures in modern artificial intelligence. He was co-founder and chief scientist of OpenAI from 2015 to 2024, and in June 2024 founded Safe Superintelligence Inc. (SSI), a company focused exclusively on building safe superintelligent AI.

Sutskever is widely credited as a key architect of the deep learning revolution. His contributions span the AlexNet breakthrough in computer vision, the sequence-to-sequence framework for neural machine translation, and the research direction behind the GPT series of large language models.

Early life and education

Sutskever was born in Nizhny Novgorod (then Gorky), Russia, in 1986. His family emigrated to Israel when he was a child, and he later moved to Canada. He studied mathematics and computer science at the University of Toronto, where he began working with Geoffrey Hinton.

Sutskever completed his PhD under Hinton's supervision at the University of Toronto in 2013. His doctoral work focused on training recurrent and deep neural networks, and on understanding the optimisation landscape of deep models.

Career

AlexNet (2012)

While still a PhD student, Sutskever was one of the three authors — with Alex Krizhevsky and Geoffrey Hinton — of the AlexNet paper ("ImageNet Classification with Deep Convolutional Neural Networks", NeurIPS 2012). AlexNet won the ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2012 by a dramatic margin, reducing the top-5 error rate from 26% to 16%. The result is widely regarded as the catalyst for the modern deep learning era.

AlexNet demonstrated that deep convolutional neural networks trained on GPUs could vastly outperform hand-engineered feature extractors on large-scale image recognition, and triggered an industry-wide shift toward deep learning.

Sequence-to-sequence learning (2014)

In 2014, Sutskever, Oriol Vinyals, and Quoc Le published "Sequence to Sequence Learning with Neural Networks" (NeurIPS 2014), which introduced the encoder-decoder framework for mapping variable-length input sequences to variable-length output sequences using LSTMs. This architecture became the foundation for neural machine translation and many subsequent natural language processing systems, and was a direct precursor to the attention mechanism (Bahdanau et al., 2014) and ultimately the Transformer.

Google Brain (2012–2015)

After completing his PhD, Sutskever spent three years at Google Brain as a research scientist, where he worked on deep learning for natural language understanding and sequence modelling. During this period he developed the sequence-to-sequence framework and contributed to advances in neural network optimisation.

OpenAI (2015–2024)

Sutskever was a co-founder of OpenAI in December 2015, alongside Sam Altman, Greg Brockman, Elon Musk, and others. He served as chief scientist from the organisation's inception.

At OpenAI, Sutskever led or oversaw much of the core technical work that produced:

GPT (2018): the first generative pre-trained transformer, demonstrating that unsupervised pre-training on large text corpora followed by supervised fine-tuning could achieve strong performance across NLP tasks.
GPT-2 (2019): a 1.5-billion-parameter language model whose outputs were considered so convincing that OpenAI initially staged its release, citing concerns about misuse.
GPT-3 (2020): a 175-billion-parameter model that demonstrated surprising few-shot learning abilities and launched the era of prompt engineering.
GPT-4 (2023): a multimodal model widely reported to use a mixture-of-experts architecture with approximately 1.76 trillion parameters.

Sutskever is known for his conviction, expressed as early as 2015–2016, that scaling up language models with more data and compute would yield qualitatively new capabilities — a view that was vindicated by the scaling laws literature and by GPT-3/4's emergent abilities.

November 2023 board crisis

On 17 November 2023, OpenAI's board of directors fired Sam Altman as CEO, citing a loss of confidence in his candour. Sutskever was reported to have initially supported the board's decision. The firing triggered a staff revolt: over 700 of OpenAI's ~770 employees signed a letter threatening to leave for Microsoft unless the board resigned and reinstated Altman. Within days, Altman was reinstated as CEO and the board was reconstituted. Sutskever subsequently expressed regret over the episode.

Departure

On 14 May 2024, Sutskever announced his departure from OpenAI, posting on social media: "I'm confident that OpenAI will build AGI that is both safe and beneficial." His departure was widely interpreted as connected to disagreements about the balance between safety research and commercial deployment.

Safe Superintelligence Inc. (2024–present)

In June 2024, Sutskever co-founded Safe Superintelligence Inc. (SSI) with Daniel Gross and Daniel Levy. The company's stated mission is to build safe superintelligence — and nothing else — treating safety and capabilities as inseparable engineering problems. SSI raised $1 billion in September 2024 at a $5 billion valuation, despite having no product and no revenue.

SSI is headquartered in Palo Alto, California, with a research office in Tel Aviv, Israel. Sutskever has stated that the company intentionally avoids the pressures of products, revenue, and customer management in order to focus on its core objective.

Views on AI

Sutskever has been vocal about several themes:

Scaling is key: he consistently argued that scaling up models, data, and compute would produce capabilities that smaller models could not exhibit, well before this became the consensus view.
AI safety as an engineering problem: at SSI, he has framed safety not as an external constraint on capability but as a core technical challenge to be solved alongside capability development.
Superintelligence is near: Sutskever has expressed the belief that artificial superintelligence may be achievable within the current decade, and that ensuring its safety is the most important technical problem of the era.
Compression as understanding: he has articulated the view that a sufficiently powerful predictive model (i.e. one that compresses data well) necessarily develops genuine understanding of the world, challenging the "stochastic parrot" critique of large language models.

Awards and recognition

NeurIPS Test of Time Award (2022, for the AlexNet paper)
Fellow of the Royal Society of Canada
Named one of Time magazine's 100 Most Influential People in AI (2023)
Cited in the 2024 Nobel Prize in Physics, awarded to Geoffrey Hinton and John Hopfield for foundational work on artificial neural networks — work Sutskever directly extended

References

Krizhevsky, A.; Sutskever, I.; Hinton, G. E. (2012). "ImageNet Classification with Deep Convolutional Neural Networks". NeurIPS 2012.
Sutskever, I.; Vinyals, O.; Le, Q. V. (2014). "Sequence to Sequence Learning with Neural Networks". NeurIPS 2014. arXiv:1409.3215.
Radford, A.; Narasimhan, K.; Salimans, T.; Sutskever, I. (2018). "Improving Language Understanding by Generative Pre-Training". OpenAI.
Brown, T. B. et al. (2020). "Language Models are Few-Shot Learners". NeurIPS 2020. arXiv:2005.14165.
"OpenAI's Chief Scientist Is Leaving". The New York Times. 14 May 2024.
"Ilya Sutskever Launches Safe Superintelligence Inc.". Bloomberg. 19 June 2024.
"Safe Superintelligence Raises $1 Billion". Reuters. 4 September 2024.