In artificial intelligence circles, the word "recursion" has become a new trend. Two separate startups have already chosen this name, and many companies are mentioning Recursive Self-Improvement (RSI) technology in their roadmaps. Like AGI, RSI has become a symbol for a sharp leap in artificial intelligence, although there are still different views on its exact definition. This is reported by Techcrunch.com reports .
In simple terms, RSI is an artificial intelligence system capable of continuous self-updating. If systems start managing this process better than humans, the human factor may become redundant. Renowned researcher Richard Socher recently launched the Recursive Superintelligence project for this very purpose. According to him, the processes of formulating, implementing, and verifying research ideas will be fully automated.
Legendary expert Andrej Karpathy, who worked at companies like Tesla and OpenAI, is also working on the Auto-Research project. He is training LLM models for simple tasks using a swarm of agents. Currently, Karpathy is working at Anthropic, where he will have the opportunity to apply his ideas on a larger scale.
Also, the Adaption startup, founded by former Google and Cohere employee Sara Hooker, introduced the AutoScientist tool. This system trains agents to implement small-scale improvements. Doris Xin, founder of Disarray, noted that her self-teaching agent won 28 medals in Kaggle competitions, emphasizing that the main problem is not creativity, but engineering reliability.
However, there is plenty of evidence that the artificial intelligence industry is still far from true recursive systems. Google CEO Sundar Pichai also recently admitted in an interview that balancing progress in this field with public concerns is difficult.
Read “Zamin” on Telegram!
