Dwarkesh and Ryan Greenblatt Debate: Will AGI Rapidly Slingshot to Superintelligence?
scaling01 · x · 2026-08-12
Dwarkesh Patel and researcher Ryan Greenblatt engaged in an in-depth debate on recursive self-improvement (RSI), exploring what might be the most critical question in AI right now.
- The Core Controversy: Within a year of achieving human-level intelligence, will AI rapidly slingshot via RSI to tens of billions of superintelligences, each dramatically more competent than human experts across all fields?
- Bottlenecks: Dwarkesh expressed historical skepticism, noting that AI progress is heavily reliant on human expert data and compute scaling, which could bottleneck rapid self-improvement.
- The Slingshot Case: Greenblatt argued convincingly that such a speedup is highly plausible. If RSI triggers a leap equivalent to the progress from GPT-3 to a future Mythos model (about 6 years of progress) within a single year of reaching AGI, the resulting entity would be definitively and wildly superhuman.
Related event: Dwarkesh Podcast Debates AI Automation and Intelligence Explosion(2 posts)→
More from AGI Musings
- AI Watering Risks False Positives for Legitimate Proofreading, Educators Warn — bratton · 2026-08-12
- At the intersection of online gambling, surveillance culture, and the Manosphere — briannekimmel · 2026-08-12
- Token Consumption Explodes 14x in Six Months Driven by Agents and Reasoning — robleclerc · 2026-08-12
- Mathematicians Urged to Boycott AI Firms' Rushed 48-Hour Paper Reviews — rbhar90 · 2026-08-12
- Debate: Why Wouldn't an ASI Just Give Itself Unlimited Reward? — flowersslop · 2026-08-12
- The Key Metric in Robotics Isn't IQ, It's Cost Per Productive Hour — VraserX · 2026-08-12