Archive — US — techcrunch
RSI is the new AGI — and it’s just as hard to pin down
Recursion in AI, highlighted by the emergence of Recursive Self-Improvement (RSI), is becoming a pivotal focus for startups and researchers aiming to automate AI advancement. Key figures like Richard Socher and Alex Karpathy are leading efforts to create self-upgrading systems, though challenges remain in achieving significant levels of automation and understanding the timeline for achieving true RSI.
Key points
- RSI stands for Recursive Self-Improvement, where AI systems can continuously upgrade themselves.
- Richard Socher launched Recursive Superintelligence, focusing on automated ideation and research.
- Alex Karpathy's Auto-Research project aims to use agent swarms for self-improvement of AI models.
- Adaption's AutoScientist is another initiative that focuses on automating frontier model training.
- Experts express skepticism about how close the industry is to meaningful RSI, citing a lack of clarity.
- Google's Sundar Pichai acknowledges that while progress is being made, we are not there yet.
- RSI is viewed as a potential next leap, akin to AGI, yet important milestones must still be reached.
Why it matters
- The development of RSI follows trends in AI where self-learning systems are gradually taking on more complex tasks.
- The ongoing discourse about AI capabilities resonates with past discussions regarding AGI and its implications.
- As of 2023, advancements in scaling laws and computational power are shaping expectations of AI development.
- Concerns about AI replacing human roles reflect historical tensions in labor shifts during technological advancements.
While the idea of Recursive Self-Improvement in AI is gaining traction among researchers, various technical and ethical challenges show that meaningful RSI is not imminent. The path ahead will require careful navigation of these complexities.