'A billionaire's chief of staff': How this AI pioneer puts a dozen agents to work

Illia Polosukhin, a notable figure in the field of artificial intelligence, utilizes a team of AI agents to assist him with various tasks on a daily basis. One of these tasks includes summarizing meeting notes, Google Drive documents, and Slack messages to provide him with insights that enhance his performance as a CEO. He describes this system as akin to having a 'billionaire's chief of staff' at his disposal, which he employs on a weekly basis to better understand his challenges and decision-making processes. Polosukhin's vision extends beyond personal productivity; he anticipates a future where AI agents will play a significant role in the global economy, executing trades, coordinating supply chains, and facilitating transactions for individuals and businesses alike. However, he warns that society is unprepared for the implications of artificial general intelligence (AGI) and the transformation it brings to various systems, including the internet and government institutions. As a coauthor of the influential 2017 paper 'Attention Is All You Need,' which introduced the Transformer architecture pivotal to modern AI, Polosukhin has been at the forefront of AI development. He founded NEAR AI with the goal of enabling machines to generate software through natural language commands, a concept that seemed far-fetched at the time but is now gaining traction under the term 'vibe coding.' Polosukhin is not surprised by the rapid advancements in AI capabilities, including those demonstrated by models like Anthropicon's recent preview, Mythos, which are reportedly capable of identifying and exploiting vulnerabilities. He cautions that as AI models evolve, they often present new risks, describing the process as a continuous 'cat and mouse' game. To mitigate these risks, Polosukhin emphasizes the need for a robust trust and security framework within AI systems. At NEAR, he is working on creating an infrastructure that reduces reliance on any single AI lab, thereby safeguarding users' data and ensuring that AI agents operate transparently. He argues that as AI systems handle sensitive information, it is crucial to prevent any one entity from having monopolistic control over this data. Another concern he addresses is the potential for manipulation of information provided by AI. As users increasingly rely on AI for news and investment guidance, there is a risk that biased or malicious alterations could shape the output. He cites an incident from xAI, where unauthorized modifications resulted in problematic responses from its AI, highlighting the necessity for an open-source, auditable AI platform that enhances user oversight. Currently, while Polosukhin's AI agents offer valuable assistance, he remains cautious about their reliability. He shared how one agent can compile news regarding geopolitical events and assess market impacts, while others assist with coding and business growth strategies. Despite their utility, he emphasizes the importance of maintaining human oversight, as AI still requires careful management to ensure sound decision-making. 'If I let it run unchecked, I often end up with nonsensical outcomes,' he noted, underscoring the need for human judgment in AI operations.