Illia Polosukhin
Last reviewed
Jun 9, 2026
Sources
22 citations
Review status
Source-backed
Revision
v3 · 1,869 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Last reviewed
Jun 9, 2026
Sources
22 citations
Review status
Source-backed
Revision
v3 · 1,869 words
Add missing citations, update stale details, or suggest a clearer explanation.
Illia Polosukhin is a Ukrainian computer scientist and entrepreneur best known as one of the eight co-authors of the 2017 paper "Attention Is All You Need", which introduced the Transformer architecture that underpins modern large language models. [1] He spent roughly three years at Google Research before co-founding NEAR, a startup that began as a machine-learning company under the name NEAR.ai and pivoted into the NEAR Protocol blockchain. [2][3] As of the mid-2020s he leads NEAR AI, an open research effort aimed at user-owned artificial intelligence. [4]
Polosukhin's career spans two distinct waves of computing. In the first, as an engineer at Google Research, he worked on TensorFlow and natural language understanding and contributed to the Transformer paper, work that helped reshape deep learning for sequence modeling. [2][3] In the second, he co-founded NEAR, a company that set out to teach machines to write code, hit practical limits in coordinating and paying a distributed workforce, and reoriented around building a layer-1 blockchain to solve that coordination problem. [3][5] His recent work returns to artificial intelligence, framing blockchain primitives as infrastructure for open, privacy-preserving AI. [4][6]
According to his Google Scholar profile, which lists NEAR as his affiliation and a verified near.ai email address, his publications had been cited more than 250,000 times, the overwhelming majority of which trace to "Attention Is All You Need." [7] The same profile reports an h-index of 41 and an i10-index of 44, with "Attention Is All You Need" alone accounting for more than a quarter of a million citations. [7]
Polosukhin is from Ukraine and grew up in Kharkiv. [3][8] He has said his interest in machine learning and artificial intelligence dates to childhood, crediting films such as "The Matrix" and "A.I." as early influences. [3] He took up programming young and competed in programming contests during high school and at university, an experience he later described as planting the seed for NEAR's original idea of automating the work of software engineers. [14][15]
He earned a master's degree in applied mathematics and computer science from the National Technical University "Kharkiv Polytechnic Institute," where he also completed his undergraduate studies. [8][14][15] He relocated to the United States, settling in California, after finishing his degree. [8][14]
Polosukhin joined Salford Systems, a predictive-analytics and data-mining software company (later acquired by Minitab), as a software developer in 2008, where he worked for roughly six years. [8][15] There he built tools for big-data predictive analytics, text mining, and geo mining, and helped refine the company's Salford Predictive Miner toolkit. [8][15] In January 2014 he joined Google Research, where he rose to an engineering management role within his first year. [8][9]
At Google, Polosukhin managed a team of researchers focused on deep learning and natural language understanding, and worked on TensorFlow, the company's open-source machine-learning framework. [2][9][15] His TensorFlow work included the high-level learning interface that was known as SKFlow and later TF.Learn. [15] Much of his applied research targeted question answering for Google Search, aimed at returning precise answers to user queries. [9][14]
He co-authored several research papers during this period. Beyond the Transformer paper, his most-cited works include the "Natural Questions" benchmark, a question-answering dataset built from real, anonymized Google Search queries published in the Transactions of the Association for Computational Linguistics in 2019, and the "WikiReading" large-scale language-understanding task over Wikipedia presented in 2016; he also co-authored work on coarse-to-fine question answering for long documents in 2017. [7] He left Google in 2017. [9]
"Attention Is All You Need," submitted to arXiv on 12 June 2017 and published at the NeurIPS conference later that year, proposed the Transformer, a sequence-transduction model built entirely on attention mechanisms and dispensing with recurrence and convolution. [1] The paper states that all eight authors were equal contributors and that the listing order was randomized; Polosukhin appears as the eighth and final listed author. [1] The eight authors and their listed order are shown below.
| # | Author | Notable later affiliation |
|---|---|---|
| 1 | Ashish Vaswani | Co-founder, Essential AI |
| 2 | Noam Shazeer | Co-founder, Character.AI; later Google |
| 3 | Niki Parmar | Co-founder, Essential AI |
| 4 | Jakob Uszkoreit | Co-founder, Inceptive |
| 5 | Llion Jones | Co-founder, Sakana AI |
| 6 | Aidan N. Gomez | Co-founder and CEO, Cohere |
| 7 | Łukasz Kaiser | OpenAI |
| 8 | Illia Polosukhin | Co-founder, NEAR |
The paper's author-contributions footnote credits Polosukhin alongside Vaswani with the earliest implementation work, stating that "Ashish, with Illia, designed and implemented the first Transformer models and has been crucially involved in every aspect of this work." [10] His work centered on the machine-translation experiments that first demonstrated the architecture's value before it was generalized to other language tasks. [2][9] On the WMT 2014 English-to-French translation task, the Transformer set a new single-model state of the art while training in a small fraction of the time required by earlier leading systems. [1]
In March 2024, seven of the eight authors gathered as a group for the first time for an onstage conversation with NVIDIA chief executive Jensen Huang at the company's GTC conference in San Jose, one of the event's most heavily attended sessions; Polosukhin took part, while Niki Parmar was unable to attend. [16][17]
In 2017 Polosukhin co-founded NEAR.ai with Alexander Skidanov, a former database engineer, and served as the company's chief technology officer. [3][8][14] The company was initially a machine-learning startup with no connection to blockchain; its aim was program synthesis, teaching machines to write computer code from natural-language descriptions, with the broader ambition of working toward general-purpose AI. [3][9][18] To gather training data, the team paid computer-science students around the world through a crowdsourcing system, and ran into difficulty making reliable international payments through conventional financial rails such as PayPal and Wise, which worked poorly in countries including Ukraine and China. [3][18] That coordination-and-payments problem led the founders to study existing blockchains and then build their own. Polosukhin has described the shift as abrupt, saying the company went "from three people doing AI to nine people doing blockchain in the span of a week." [3]
The result was NEAR Protocol, a proof-of-stake layer-1 blockchain designed to scale to large numbers of users without spiking fees, to simplify onboarding for non-crypto users through human-readable account names, and to let developers build using familiar languages such as JavaScript and Rust. [3][5][18] To scale throughput, NEAR uses a sharding design called Nightshade that distributes work across multiple parallel shards. [5][18] NEAR's mainnet launched in April 2020 and became community-operated later that year. [11]
The project raised capital across several rounds. It announced a $21.6 million round led by Andreessen Horowitz, with participation from investors including Pantera Capital and Electric Capital, alongside the mainnet launch in 2020, and a larger $150 million raise in January 2022. [11][19] Reported totals for the broader NEAR ecosystem run into the hundreds of millions of dollars. [5][8]
Polosukhin is a co-founder of NEAR Protocol and has remained one of the project's most public figures. [5][12] In November 2023, announced at the NEARCON conference in Lisbon, he took on the role of chief executive of the NEAR Foundation, the nonprofit that stewards the ecosystem, succeeding Marieke Flament, who had led the foundation since early 2021; Chris Donovan, previously named as her successor, moved into a chief operating officer role. [12][20] Beyond foundation governance, he has been closely associated with NEAR's technical direction and its push to position the network as infrastructure for artificial intelligence. [5][6]
Polosukhin leads NEAR AI, an open-source research effort built around the idea of "user-owned AI," in which individuals can use and benefit from powerful models without surrendering their data to a small number of proprietary providers. [4][6] The work centers on what NEAR calls Decentralized Confidential Machine Learning, a network that aims to run training, fine-tuning, and inference across GPUs and data centers worldwide without exposing user data, model parameters, or the underlying code. [13] NEAR frames the approach around four goals: privacy, so users can run models without revealing personal data; openness, so models are transparent and verifiable; sustainable economics, so model creators can monetize their work; and community collaboration, so groups can jointly develop models. [13]
In March 2025 Polosukhin presented this NEAR AI research on confidential, decentralized AI computation at NVIDIA's GTC conference in San Jose. [13] He has argued publicly that AI will increasingly mediate how people use computers, potentially displacing traditional operating systems and applications, and that blockchains can serve as a root of trust for verifying AI behavior and provenance. [6] In doing so he positions NEAR AI as a return to the company's original artificial-intelligence mission, now built on the blockchain infrastructure it created along the way. [4][6]
Following Russia's full-scale invasion of Ukraine in February 2022, Polosukhin helped create the Unchain Fund, a cryptocurrency-based relief effort that converted donations into local currency to buy and distribute goods for civilians affected by the war. [21][22] News coverage in 2022 reported the fund had gathered on the order of several million dollars in crypto donations for humanitarian aid. [21][22]