How will Recursive push self improving AI toward ASI?

TL;DR

Recursive aims to accelerate self improving AI to approach ASI by combining large language models, data, simulations, and automation to weave together multiple scientific disciplines. The project secures heavy funding from Google Ventures, Nvidia, and AWS compute, and builds toward a future with broader scientific breakthroughs enabled by AI driven research and automation.

Transcript

Richard, I'm I'm curious. Where are we at the moment with RSI >> in various weak forms? We already have RSI. We're not quite there yet, but we're very close. >> When do you believe we reach ASI? >> I think we'll take us probably several decades. >> On Terminal Bench 4.0, Sonnet 5.5 jumped from 10% to 70%. Pretty extraordinary. My theory would be th... Read More

Key Insights

• Recursive is betting on recursive self improvement as a path toward ASI and significant productivity gains in science
• The Eureka Machine argues AI will accelerate scientific discovery by connecting hypothesis, experiment, data, and theory across disciplines
• Full stack AI for science relies on world knowledge from LMS, digitized data, simulations, and robotic process automation to weave disciplines together
• A swarm of AI agents could coordinate automated research tasks, enabling rapid exploration and validation of scientific ideas
• There is optimism about timelines, with emphasis on incremental progress and multiple disciplines advancing in parallel rather than a single threshold
• The discussion acknowledges competition with open source models and the importance of compute and data availability
• Timelines for curing diseases or solving physical sciences are framed as gradual progress rather than a single leap
• Ethical and governance considerations appear as part of the broader conversation about recursive self improvement and ASI

Questions & Answers

Q: What is full stack AI for science as described in the podcast?

Full stack AI for science refers to using AI to connect and automate every stage of scientific work, from generating hypotheses to running experiments, collecting data, and deriving theories. It relies on world knowledge in large models, digitized scientific data, improved simulations, and robotic process automation. This integrated approach enables researchers to weave together disciplines that are traditionally fragmented and operate at different levels of abstraction.

Q: Why do the speakers believe progress in science has slowed, and how might AI help?

The speakers argue progress has slowed due to fragmentation across many sub disciplines and niches, which makes cross collaboration difficult. AI can help by providing a common framework and tools that bridge disciplines, enabling data and insights to flow between fields, and by coordinating large scale automated experimentation and analysis, potentially accelerating breakthroughs across biology, physics, and computer science.

Q: What evidence of funding and partnerships supports Recursive’s approach?

The episode notes a $670 million fundraise from Google Ventures, Graycraft, Nvidia, and AMD, plus $410 million in compute from AWS. These investments, along with speaker affiliations and the mention of a book release and AI benchmarks, illustrate broad industry support for recursive self improving AI and its potential to scale scientific discovery.

Q: How is the concept of ASI addressed in the discussion?

The discussion frames ASI as the eventual outcome of recursive self improvement and argues it will emerge through coordinated AI agents and evolving capabilities rather than a single abrupt threshold. The participants emphasize timelines as gradual and contingent on compute, data, and governance considerations, rather than predicting an instantaneous leap.

Q: What role do other tech leaders play in this vision, according to the podcast?

Tech leaders are described as co-funders, collaborators, or competitors in the race toward advanced AI. Elon Musk and Palmer Lucky are mentioned in relation to Pentagon projects, while Google, Nvidia, AMD, and AWS are named as financiers and infrastructure providers, illustrating how industry players contribute to and influence the development of recursive, self improving AI.

Q: What is the Eureka Machine book arguing about AI in science?

The Eureka Machine argues that AI will deliver a century of scientific breakthroughs in the next decade by making science more programmable and interconnected. It claims that the full stack AI approach will transform hypothesis, experimentation, data handling, and theory formation across disciplines, enabling faster and more reliable innovation.

Q: What concerns or debates are raised about recursive self improvement?

The discussion acknowledges concerns about ethics and risks, and explores what restrictions or governance might be appropriate for recursive self improving AI. It also contemplates the balance between rapid advancement and safety, as well as how to manage competition with open source models while maintaining responsible development.

Q: How do the guests view the relationship between computation and scientific progress?

They view compute as a critical resource that amplifies scientific progress by enabling larger models, faster simulations, and more extensive data processing. With access to massive compute, AI systems can generalize across disciplines, automate tasks, and accelerate discovery, contributing to a broader acceleration of scientific breakthroughs.

Summary & Key Takeaways

• Recursive seeks to automate the full scientific stack from hypothesis to experiment to data and theory, enabling cross discipline progress through AI driven synthesis and robotics

• The conversation highlights large fundraisings, partnerships with major tech players, and the vision of a swarm of agents driving recursive self improvement and accelerated scientific discovery

• The discussion touches on timelines for ASI, the role of openness and competition with open source models, and forthcoming AI benchmarks

• The book The Eureka Machine argues AI will unlock a century of breakthroughs, tying together biology, physics, and computer science with programmable engineering science