# Opinion: I remain unconvinced that true recursive self‑improvement will happen soon

Digest AI · Research · published 2026-09-19T15:42:20Z

Canonical: https://digestai.news/story/opinion-i-remain-unconvinced-that-true-recursive-selfimprovement-will

## Summary

The author argues that the hype around rapid recursive self‑improvement (RSI) is overstated. He notes that only two frontier labs—OpenAI and Anthropic—are deploying thousands of concurrent agents, and that this culture amplifies anxiety about AI risks. Citing recent podcasts with Noam Brown, John Schulman, Beren Millidge and Charlie O’Neill, he highlights varied timelines: about a year for AI to act as a remote white‑collar worker, 2‑3 years for broader generality, and 5‑10 years for full superintelligence. The piece stresses that current techniques solve well‑defined problems but do not produce a magical leap in generalization, and that scaling laws still demand exponential compute for linear intelligence gains.

Internal system cards for Claude Fable 5.1 and Mythos 5.1 acknowledge AI agents help maintain progress but show no clear sign of dramatic acceleration. The author concludes that RSI may make LLMs cheaper and improve inference‑time scaling, yet the hardest barrier—peak intelligence—remains resistant, making extinction‑risk narratives premature.

## Key points

- OpenAI and Anthropic run thousands of AI agents internally, fueling both productivity and risk anxiety.
- Podcast guests estimate broad white‑collar AI capability in ~1‑2 years, full generality in 3‑5 years, superintelligence in 5‑10 years.
- Claude Fable 5.1 and Mythos 5.1 system cards say AI usage sustains progress but shows no dramatic acceleration.

## Why it matters

Clarifying the realistic limits of recursive self‑improvement tempers hype, guides investment decisions, and informs policy debates about AI safety and timelines.

## Sources

1. [Why I still haven’t bought into true RSI](https://interconnects.ai/p/where-i-stand-on-rsi) (Interconnects, 2026-09-19)

Part of the developing story: [AI Agents Tackle Millennium Prize Problems](https://digestai.news/thread/openai-reportedly-solved-a-millennium-prize-problem-using-10-000-agents) (2 stories)

## Cite

Digest AI, "Opinion: I remain unconvinced that true recursive self‑improvement will happen soon", 19 September 2026, https://digestai.news/story/opinion-i-remain-unconvinced-that-true-recursive-selfimprovement-will

---

Written by Digest AI's editorial model from the linked sources; the sources are the record. Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse
JSON: https://digestai.news/story/opinion-i-remain-unconvinced-that-true-recursive-selfimprovement-will.json