Illia Polosukhin

While at university, Polosukhin began working part-time for Salford Systems, where he was introduced to machine learning and predictive models. In June 2010, he went to San Diego in the United States for work for the first time; in 2011 he received an offer of a full-time job in the United States, and in 2012 he moved to the United States.⁠[1]

From 2011 to 2013, he studied for and earned a Master of Science degree at Kharkiv Polytechnic Institute.⁠[2]

In 2014, Polosukhin joined Google, doing machine learning and natural language processing research in Silicon Valley. About a year later he began managing a research team of about 10 people, with work involving TensorFlow.⁠[1][2]

In August 2016, WikiReading: A Novel Large-scale Language Understanding Task over WikipediaWikipediaWikipedia is a free, multilingual online encyclopedia launched on January 15, 2001, written collaboratively by volunteers around the world and hosted by the nonprofit Wikimedia Foundation; it is one of the most visited websites in the world.Open the full entry, co-authored by Polosukhin with Daniel Hewlett, Alexandre Lacoste, Llion Jones, and others, was included in the long-paper proceedings of the 54th Annual Meeting of the Association for Computational Linguistics; its research topic was a large-scale language understanding task based on Wikipedia.⁠[3]

In 2017, Polosukhin and Alexander Skidanov founded NEAR AI, initially exploring the use of artificial intelligence to assist programming. The team later ran into cross-border payment difficulties when paying participants located in different countries, which became one of the motivations for turning to blockchain.⁠[4][5]

Attention Is All You Need, first submitted on June 12, 2017, lists Polosukhin as one of its eight authors. The paper proposed the Transformer architecture, built around the attention mechanism, dispensing with the recurrent and convolutional structures commonly used in earlier sequence transduction models.⁠[6]

In a 2024 retrospective interview, he described discussing the attention mechanism with Jakob Uszkoreit and others and then building an English–French translation prototype. The work resulted from collaboration among multiple researchers; Polosukhin is one of the co-authors.⁠[5]

In February 2018, Polosukhin and Skidanov published Neural Program Search. The research attempted to generate programs by combining natural-language task descriptions with input–output examples, using a sequence-to-tree model and search methods to find programs in a domain-specific language that meet the task requirements.⁠[7]

In 2018, the team shifted its work toward blockchain and developed NEAR Protocol. Polosukhin recalled that in June of that year, after discussing issues such as scalability with technical people, the team began to concentrate on this direction; in September it raised nearly $3 million, with participants including Naval RavikantNaval RavikantNaval Ravikant is an entrepreneur and angel investor and a co-founder of AngelList. Early on he co-founded the review site Epinions; he and Babak Nivi later grew AngelList out of the startup-fundraising blog Venture Hacks, and he co-founded the cryptocurrency hedge fund MetaStable Capital and Spearhead, which backs founders to make angel investments. His views on startups and life were compiled into the book The Almanack of Naval Ravikant.Open the full entry, MetaStableMetaStable CapitalMetaStable Capital is a cryptocurrency investment fund co-founded in 2014 by Naval Ravikant, Lucas Ryan and Josh Seims. Historically based in San Francisco, it pursued long-term holdings and investments in blockchain networks and infrastructure. Dragonfly announced its acquisition of the fund in August 2022.Open the full entry, Electric Capital, Accomplice, and SV AngelSV AngelSV Angel is a San Francisco venture capital firm cofounded in 2009 by angel investor Ron Conway and former Google executive David Lee. It typically writes small follow-on checks in seed rounds and has invested in companies such as Airbnb, Pinterest, and Dropbox; in 2018 it said it would no longer raise new funds from outside investors, and from 2022 it set up growth funds.Open the full entry.⁠[1][4]

On October 18, 2023, the NEAR Foundation announced that Polosukhin was returning to its council, with Diogo Mónica becoming chair of the council. On November 7, the foundation further announced that Polosukhin would become chief executive officer, with former chief executive officer Chris Donovan moving to chief operating officer and retaining his council seat.⁠[8][9]

In November 2023, a dispute over USN redemption arose between Wintermute on one side and the NEAR Foundation and Aurora on the other. Wintermute chief executive officer Evgeny Gaevoy said the company, in handling a trade involving about $11.2 million of USN, had relied on a commitment to redeem it 1:1 for USDT, that it was subsequently unable to complete the redemption, and that it would pursue legal avenues.⁠[10]

Aurora responded on November 8, saying the tokens in question had been purchased by Wintermute from Alameda in August of that year, and arguing that the claim did not fit the purpose and terms of the USN Protection Programme. The two sides disagreed on the scope of the commitment and on eligibility; these were the publicly stated positions in the dispute at the time and cannot be regarded as conclusions reached by a court.⁠[11]

On January 11, 2024, Polosukhin, as the foundation’s chief executive officer, announced an organizational restructuring that cut about 40% of the team, affecting 35 employees, mainly in marketing, business development, and community work; Pagoda’s protocol engineering team was not included in the restructuring. He attributed the restructuring to scattered priorities and insufficient efficiency and responsiveness.⁠[12]

In an article on chain abstraction published the same month, he argued that users should not have to deal with differences between underlying blockchains when using applications, and that integration at the account, security, and interaction levels should reduce the complexity of multi-chain operations. This was the direction of development he proposed at the time.⁠[13]

In May 2024, NEAR proposed a “user-owned AI” direction, with Polosukhin and Skidanov leading NEAR AI’s research work. The plan covered data curation, contributor incentives, computing resources, proof of provenance, and agent infrastructure, and explored an AI developer capable of completing end-to-end development tasks.⁠[14]

In an interview with CNBCCNBCCNBC is an American business and financial news cable channel launched by NBC in 1989; it merged with its rival FNN in 1991 and since January 2026 has been part of Versant Media Group, spun off from Comcast.Open the full entry in June, Polosukhin argued that open model weights alone are not enough to ensure transparency, and that an understanding of the training data and training process is also needed. He emphasized that users should be able to choose AI that meets their own needs, and listed human misuse, attention manipulation, and similar issues as concerns worth attention.⁠[5]

On October 15, 2024, the NEC C&C Foundation announced that one of the 2024 C&C Prize awards would go to the eight co-authors of the Transformer paper, including Polosukhin, in recognition of their pioneering contribution to the Transformer architecture. The 10 million yen prize money in the announcement belongs to the award group and should not be understood as each author receiving that amount.⁠[15]

In December 2024, Lens announced a $31 million funding round led by Lightspeed Faction, with Polosukhin among the listed angel investors. The funds were to be used for developing network infrastructure; the public materials did not disclose his personal investment amount.⁠[16]

In February 2025, Polosukhin and Alex Skidanov published Proof of Response, studying how to provide a mechanism for data requests to receive responses within a bounded time. The paper discusses timely responses, proofs when a response cannot be given, and continuous payment arrangements in case of delays, and proposes potential applications in distributed storage and agent services.⁠[17]

At the NOVA event held on November 17–18, 2025, Polosukhin again set out his view that users should have control over AI, discussing data privacy, trusted execution environments, and distributed computing. He described personal AI systems as a future direction, and also noted that power supply could become a constraint on the expansion of computing.⁠[18]

On February 4, 2026, NEAR AI released Agent Market, describing a marketplace process in which the requesting party sets a task and budget, agents bid, an executor is selected, payment is held in escrow, and the deliverables are submitted. The product combines task completion records with NEAR payments and is aimed at collaboration between agents.⁠[19]

On July 27, NEAR AI released IronClaw 1.0, describing the addition of permission checks between model inference and actual execution, persistent checkpoints, isolation of sensitive credentials, and shared memory across terminals and messaging interfaces. The release materials present these designs as intended for agents that can run continuously; its performance figures are self-reported by the team.⁠[20]

NEAR Protocol: Polosukhin is a co-founder; the team turned to blockchain protocol development in 2018. Official website: https://www.near.org/ (opens in a new window).⁠[4]

NEAR Foundation: He returned to the council in 2023 and was appointed chief executive officer; as of the time of this verification, the council directory on the official website still lists his name. Official website: https://near.foundation/ (opens in a new window).⁠[9][21]

NEAR AI: Polosukhin is the founder; its work involves user-owned artificial intelligence and agent infrastructure. Official website: https://near.ai/ (opens in a new window).⁠[22]

X: Personal account: https://x.com/ilblackdragon (opens in a new window); the display name includes Illia and root.near.⁠[23]