Жуткое будущее ИИ | Варламов и Дороничев в AI-столице мира Кремниевой долине

Summary

In this interview, journalist Ilya Varlamov speaks with AI entrepreneur and former Google manager Andrey Doronichev about the rapid development of artificial intelligence, its effects on work and society, and the risks of increasingly capable systems.

Key technological developments

• AI progress and recursive self-improvement: Doronichev argues that AI is moving beyond being a helpful “genius in a box.” Leading labs are trying to build systems that can help develop more capable versions of themselves, a process often called recursive self-improvement. If it works, progress could accelerate sharply.
• AI-assisted programming: He sees coding as one of the most significant current advances. He describes Claude Code as an example of an AI coding agent that can work on tasks for hours, with people increasingly acting as managers who specify goals, supervise agents, and review their output. He claims that a recent version of Claude Code was largely written with its own assistance.
• Biotechnology and drug discovery: Doronichev discusses AI tools for predicting protein structures and designing drug molecules. He says the next major bottleneck is testing candidates: experiments with cells still take time and cannot yet be fully replaced by simulation. More realistic computer models of cells, organs, and eventually people could speed up the research cycle. He also mentions AI-assisted research into a molecule for a rare genetic disease and cites developments in treatments for hepatitis C, HIV, and cancer as examples of medical progress.
• AI beyond chatbots: The discussion covers coding agents, protein-folding models, AI-driven drug discovery, and the use of AI in military systems, including drones. The speakers stress that AI’s real-world impact depends not only on what models can do, but also on how quickly organizations adopt them.

Work, education, and social change

Doronichev expects AI to substantially change or eliminate many existing jobs, but says the broader shift may be more important than individual job losses: access to skilled intellectual work could become far more abundant. He suggests that success will increasingly depend on adapting quickly rather than mastering one narrow, stable profession.

For students and workers preparing for that future, he recommends:

• Building strong foundations in subjects such as mathematics, physics, and philosophy.
• Learning new tools continuously and becoming comfortable changing direction.
• Developing independent judgment and the ability to analyze information.
• Thinking about personal motivation and purpose, rather than relying only on external pressure to work or study.

He also questions the assumption that easy access to AI will automatically make people less capable. In his view, technology creates the opportunity to disengage, but whether people do so depends on their own choices and habits.

AI companies and competition

Doronichev offers his personal assessment of the major companies:

• Anthropic: He regards it as a current leader, particularly praising its focus on AI safety and the popularity of Claude and Claude Code.
• OpenAI: He describes it as a strong competitor but says its broad ambitions and organizational changes can make it appear less focused.
• Google: He highlights its long-standing investment in AI infrastructure and its full-stack capabilities, including custom chips and models.
• Meta: He argues that its strengths lie in social platforms and advertising, and that it may be better served by concentrating on those areas.
• xAI/Grok: He believes it has struggled to retain key talent and build a strong research culture.
• Apple: He suggests that its emphasis on consumer devices, rather than large-scale AI services and server infrastructure, contributed to its slower start.
• China: He describes the United States and China as the main AI powers, with China’s ability to scale infrastructure as a major advantage. He says Russia lacks the computing resources to compete at the same level.

These are the guest’s opinions, not an independent ranking or review.

Safety, misuse, and military applications

Doronichev is more concerned about people using powerful AI systems for harmful purposes than about AI independently deciding to attack humanity. Potential risks discussed include cyberattacks, the creation of dangerous biological agents, and attacks on critical infrastructure.

He argues that simple instructions or “guardrails” may not be enough to prevent misuse. The interview discusses deeper safety techniques intended to limit what a model can do, alongside the need for oversight, regulation, and cooperation between governments. Doronichev suggests that international agreements and monitoring of computing resources may eventually be needed to limit the speed of development.

The speakers also discuss AI’s role in modern warfare, especially drones. Doronichev presents the war in Ukraine as an example of how inexpensive, software-enabled systems are changing military tactics, while noting that attacks on infrastructure are also constrained by the design and connectivity of the infrastructure itself.

Information, authenticity, and future scenarios

The interview considers how AI-generated images, video, and text could make it harder to distinguish authentic material from fabrication. Doronichev doubts that people can rely on an automatic method to separate truth from falsehood in every case. He suggests that verification tools may become more common, but distinguishes factual authenticity from the way people interpret events.

He also reflects that the internet may have passed a period when information could travel quickly while remaining relatively easy to verify. As a possible response, he imagines people relying more on trusted records and pre-AI sources.

For a five-year outlook, he sketches two broad possibilities:

• A positive outcome: AI systems become widely available assistants for education, health care, legal work, and creative tasks, while robots take on more physical work. This depends on society retaining control over the most powerful systems and distributing the benefits broadly.
• A severe outcome: Powerful tools fall into the hands of malicious actors and enable large-scale cyber, biological, or other attacks. He considers this more plausible than a scenario in which AI spontaneously becomes hostile on its own.

Reviews, guides, and tutorials

There is no formal product review or step-by-step tutorial in the interview. It includes discussion and comparisons of AI products and systems, especially Claude, Claude Code, OpenAI models, Google’s AI work, Meta, Grok, and DeepSeek. Sponsor advertisements and channel promotions are not included in this summary.

Main speakers and sources

• Ilya Varlamov — interviewer and host.
• Andrey Doronichev — interview guest; AI entrepreneur and former Google manager. The technological assessments, company comparisons, and forecasts above are primarily his views.