Dwarkesh Podcast: Will Automated AI Research Trigger an Intelligence Explosion?
Dwarkesh Patel · youtube · 2026-08-12
Dwarkesh Patel and Ryan Greenblatt dive deep into recursive self-improvement (RSI) and what happens once AI can automate AI research.
- Feasibility of Intelligence Explosion: Dwarkesh was historically skeptical of a rapid slingshot to superintelligence, citing compute and human data bottlenecks. However, Greenblatt makes a compelling case that within a single year of achieving AGI, we could see a leap equivalent to 6 years of progress, resulting in wildly superhuman AI. Greenblatt's median prediction for automating AI R&D is 2031.
- Alignment Dilemmas: They discuss who superintelligences should be aligned to. Dwarkesh worries that current specs (like the Claude Constitution) won't shape ASI into true personal advocates. They also debate whether current reward hacking extrapolates to ASIs teaming up to take over the world.
Related event: Dwarkesh Podcast Debates AI Automation and Intelligence Explosion(2 posts)→
More from AGI Musings
- RSI Simulator: A Web Game Where You Race to Build Superintelligence — AccBalanced · 2026-08-12
- The Paradox of Enterprise AI: Employees Gain Leisure, Firms Gain Nothing — arampell · 2026-08-12
- Top Quotes from Agentic AI Summit 2026: The Future is Agentic — dawnsongtweets · 2026-08-12
- UK Tenants Use AI to Generate Complaints, Shifting Bottleneck to Human Verification — rohanpaul_ai · 2026-08-12
- What People Actually Worry About Regarding AI Data Centers — iamrobotbear · 2026-08-12
- Will AI Compute Runs Dictate Global Economy? A Deep Dive into Anti-Capitalism and AI Control — jessi_cata · 2026-08-12