Summary of Key Points
RSI (Recursive Superintelligence) is a new laboratory founded by eight top AI researchers, including Daisuke Taga, former director of FAIR at Meta. The laboratory has been valued at $4.65 billion and has raised its first round of funding in the amount of $650 million. Its focus is on "AI recursive self-improvement" – enabling AI to optimize itself, complete scientific research tasks, and accelerate knowledge discovery. The article discusses RSI's technical approach, team strengths, industry competition, and potential impacts on employment. Daisuke Taga uses the metaphor of a fish trying to jump out of a drying puddle to describe the career challenges in the AI era, emphasizing that individuals need to find their unique value rather than simply playing the role of a "screw" in the larger system.
What Exactly Does RSI Do? – AI Evolving Itself Without Human Intervention
The core of RSI is "recursive self-improvement," which means using AI to replace humans in repetitive tasks involved in scientific research, such as identifying research directions, designing experiments, and optimizing models. The goal is for AI to discover new knowledge on its own. For example, while humans need to rest and eat while conducting AI research, AI can work continuously 24/7, first assisting with tedious tasks before gradually learning to find new insights on its own. Their ultimate objective is to "maximize the rate of knowledge discovery" – by providing sufficient computing resources, AI can generate new ideas and results, thereby accelerating human progress.
Here’s a concrete example: In the past, training AI models required humans to design the architecture and adjust parameters; RSI aims for AI to design experiments, evaluate outcomes, and even modify its own code and weights, becoming stronger over time. This is similar to students setting their own exam questions, correcting their homework, and improving their grades without constant supervision from teachers.
Why Do Investors Take a Risk? – A Team of Eight AI Experts Is More Reliable Than Individual Efforts
The reason investors are willing to invest $650 million in RSI lies in the expertise of its eight co-founders, all considered leaders in the AI community:
- Richard Socher (CEO): Founder of MetaMind and former chief scientist at Salesforce, with a deep understanding of both business and technology.
- Tim Rocktäschel: Former researcher at Meta/DeepMind specializing in reinforcement learning, who played a crucial role in securing funding for RSI.
- Jeff Clune: Has been dedicated to AI research for ten years, with papers published in Nature.
- Daisuke Taga: Former director of FAIR at Meta, skilled in combining modeling and engineering techniques.
Investors see the potential in RSI because AI evolves rapidly, and product directions can change at any time. However, this team is capable of responding quickly and collaborating efficiently. Moreover, as co-founders, they have a strong incentive to succeed.
Is It Safe for AI to Improve Itself? – Explainability Acts as a Brake
The biggest concern with AI evolving on its own is the risk of losing control (for example, creating a model that humans cannot manage). Daisuke Taga emphasizes that explainability is crucial:
- Security: We need to understand how AI makes decisions; for instance, if it modifies its code, we must be able to analyze the logic to assess potential risks.
- Efficiency: Explainability allows us to evaluate model performance during training, saving time and costs by avoiding the need to wait for thousands of GPUs to complete the process.
Their approach is to use humans as "referees" initially (when AI is not yet powerful enough to assess itself) and later rely on AI to verify each other’s work, ensuring that AI evolves within safe boundaries.
How Does RSI Differ from Larger Companies?
Large companies like OpenAI, Anthropic, and DeepMind are focusing on developing "coding agents" that can assist with writing code. However, RSI is pursuing a more long-term approach of "AI self-evolution":
- Focus: Large companies often spread their resources across multiple areas; RSI focuses all its funds and efforts on recursive self-improvement.
- Flexibility: Large companies make slow decisions, while small teams can experiment quickly. For example, DeepMind’s AlphaEvolve uses AI to optimize code, but RSI aims for AI to optimize itself more fundamentally.
- Cost Efficiency: Large companies rely on massive computing power; RSI seeks smarter solutions rather than simply throwing more money at the problem.
How Should We Survive in the AI Era? – Stop Being a Fish in a Drying Puddle
Daisuke Taga uses a poignant metaphor: Just as fish struggling in a drying puddle cannot survive, individuals in the AI era must adapt. He suggests that:
- AI will soon be capable of performing tasks typically done by level 4-5 engineers and will replace many repetitive jobs.
- Stop thinking of yourself as just another "screw" in a large company; instead, find your unique value – for example, engage in exploratory work (like writing novels) or share experiences that no one else has.
- The future could be like a "New Renaissance," where everyone must pursue what they truly want to do, rather than being molded into predefined roles.
In summary, AI self-evolution is a major trend, and laboratories like RSI may represent breakthroughs. However, individuals need to adapt in advance – don’t wait until the water dries up before taking action; instead, actively embrace evolution.
This interview provides a glimpse of the future of the AI industry: AI will not only replace jobs but also conduct research on its own. Our challenge is to shift from being replaced by AI to co-creating with it.