Recursive оценили в $4.65 млрд за самоулучшающийся AI с контролем безопасности

Startup Recursive has emerged from stealth with a $4.65 billion valuation. The company is developing a revolutionary approach to artificial intelligence: instead of humans training models, the system itself conducts experiments on its own improvement in a controlled environment.

New Philosophy of Machine Learning

Recursive's core idea differs fundamentally from the conventional approach to AI development. In the classical scheme, developers write algorithms, prepare datasets, and train models, step by step improving results. This is a long and resource-intensive process requiring expertise and months of work.

Recursive proposes an alternative: the system itself conducts experiments on itself. It tests different approaches to its own development, analyzes results, and selects the most effective paths for improvement. All experiments occur in a safe, isolated environment with pre-established rules and constraints.

The company particularly emphasizes that safety is not an add-on to the architecture, but an integral part from the very beginning of development. This is critically important: if a system can improve itself, then control over its actions and direction of development becomes priority number one. Recursive solves this problem not at the end of development, but embeds it in the very foundation of the system.

Leader in AI with Expertise

Founder and CEO Richard Socher is an experienced artificial intelligence specialist. He previously led the entire AI division at Salesforce, where he was responsible for integrating artificial intelligence into the company's cloud services. Socher is known for his research in natural language processing (NLP) and deep learning — fundamental areas of modern AI. His vision is to create AI that can independently and safely seek paths to its own improvement. This is an ambitious goal, and its realization requires both a significant technological breakthrough and deep understanding of safety issues in systems with a high degree of autonomy.

World-Class Investors

The $4.65 billion valuation reflects very high confidence in the idea and team. Behind Recursive stand serious players from the technology and venture capital industries:

• Google Ventures — Google's investment division
• Greycroft — a well-known venture fund specializing in innovative startups
• Nvidia Ventures — the corporate venture fund of Nvidia
• AMD Ventures — AMD's investment program

The presence of Nvidia and AMD is particularly significant — two of the world's largest semiconductor and GPU manufacturers. Their interest in self-improving AI is logical: if Recursive succeeds in creating truly effective self-improving AI, this may require significant computing power, which ultimately benefits equipment manufacturers.

Turning Point for AI

The appearance of Recursive with such a valuation shows that the AI industry stands on the threshold of a fundamental paradigm shift. From traditional machine learning, where humans completely control every step of development and training, the world is transitioning to more autonomous systems that improve themselves. But this does not mean loss of control or compromise with safety. Recursive demonstrates that safety can be built into the very architecture of the system, rather than added later as an addition. If the company and its partners succeed in proving that self-improving AI can be safely deployed in real-world conditions, this could dramatically accelerate progress in creating more efficient and intelligent systems with minimal necessary human intervention.