Are We in the Early Stages of Recursive Self-Improvement? Reddit Debates
Citadel_Employee · reddit · 2026-11-04
A Reddit thread argues we may already be in the early stages of recursive self-improvement (RSI): while humans remain in the loop, models are now good enough to generate synthetic data for training better successors, and much of the data curation pipeline can be automated. The author also cites a notable signal—Astra recently managed to edit the raw weights of Qwen 3.8—while admitting that the criteria for RSI remain as murky as those for AGI, and invites community discussion.
More from AGI Musings
- Development biologist slams AI biology breakthrough hype: 'complete bullshit' — paulnovosad · 2026-11-04
- Blogger slams Scott Alexander: opposes AI pause because he wants the Singularity — ludwigABAP · 2026-11-04
- Cambridge's David Krueger: industry surprisingly receptive to 'stop AI' talk — DavidSKrueger · 2026-11-04
- AI safety researcher David Krueger: ai4 conference audience receptive to "stop AI" concerns — DavidSKrueger · 2026-11-04
- Kevin Roose's AGI Chronicles book tour: podcast chats with Sam Harris, Steven Levy and more — kevinroose · 2026-11-04
- VC predicts AI will replace information work next year at 10,000X price cuts — davidpattersonx · 2026-11-04