Where’s the “intelligence explosion”?

Ramez Naam argues that current evidence does not support predictions of an imminent “intelligence explosion” or FOOM—a runaway cycle in which AI rapidly improves itself into artificial superintelligence. Although AI is already helping researchers and engineers, Naam estimates that the self-improvement feedback loop would need to become roughly 5–10 times stronger to sustain accelerating progress. He distinguishes five forms of recursive self-improvement, ranging from productivity gains and AI-assisted model training to fully autonomous, runaway improvement. Types 1 and 2 are already occurring, while more autonomous forms remain unproven. Naam expects autonomous self-improvement eventually, but believes a major conceptual breakthrough would be required for it to produce explosive growth. AI has already achieved narrow superhuman performance in highly verifiable domains such as games, formal mathematics, and parts of coding. However, this does not imply broad intelligence: current systems require extensive training data, learn unreliably from experience, and often fail at seemingly simple tasks. Naam emphasizes the gap between benchmark results and real-world research. OpenAI data reportedly shows models completing short research tasks autonomously only about 86% of the time, with performance declining as tasks become longer and more complex. Anthropic reports that AI participates in most internal R&D tasks, but no cases of fully autonomous completion. Overall, Naam expects rapid AI progress but sees diminishing returns, rising resource demands, and insufficient evidence of acceleration. The author remains agnostic, noting that AI could become extraordinarily capable by 2040 even without a sudden singularity.

Build this article’s timeline

Extract dated events, key actors, and open questions using the same timeline analysis available in Signal.

Log in to build

Timeline generation is available to signed-in community members.

Five Ws 🤖 3 AI prefills · 2 still unanswered
Who
Vernor Vinge, Noam Brown, Tanay Jaipuria, and OpenAI
🤖 AI prefill
What
Ramez Naam argues that current evidence does not support predictions of an imminent “intelligence explosion” or FOOM—a runaway cycle in which AI rapidly improves itself into artificial superintelligence.
🤖 AI prefill
When When did this happen, or what is the timeframe?
Where Where did this occur?
Why
Centers on recursive self-improvement (RSI), intelligence explosion, and fast takeoff / FOOM
🤖 AI prefill

Login to answer and earn XP for each gap you fill.

Recursive Self-Improvement recursive self-improvement, AI research, feedback loop, autonomy, successor models Intelligence Explosion Skepticism intelligence explosion, FOOM, fast takeoff, singularity, artificial superintelligence AI Progress Measurement benchmarks, forecasts, METR, task horizons, success rates Narrow Superintelligence narrow superintelligence, chess, Go, formal mathematics, proofs
recursive self-improvement (RSI)100% intelligence explosion98% fast takeoff / FOOM96% artificial superintelligence (ASI)94% technological Singularity93% AI self-improvement loop92% runaway superintelligence90% diminishing returns88% autonomous AI research87% narrow superintelligence86%
22
Manipulation Score: Low How much this article uses emotional framing vs. straight reporting
Loaded Language 1825%
Absolutist 1215%
Threat/Panic 2025%
Us vs Them 815%
Engagement Bait 3520%
Manipulation score look right?

Factual Claims (10)

medium technology
Ramez Naam predicted the solar and battery revolutions before they were widely understood.
medium technology
Many AI researchers, entrepreneurs, and safety advocates believe recursive self-improvement could produce a fast takeoff or technological Singularity.
medium technology
Noam Brown said recursive self-improvement and AI models conducting AI research are OpenAI's top priority by a wide margin.
low technology
AI is already helping improve AI systems, including through researcher and engineer productivity and training smaller models.
high technology
A self-improvement loop would need to be roughly 5–10 times stronger to sustain itself or become runaway.
medium technology
AI has achieved narrow superintelligence in domains such as chess, Go, formal mathematics, and parts of coding.
high scientific
OpenAI reported an AI-generated proof addressing the Navier–Stokes existence and smoothness problem.
medium technology
OpenAI models completed tasks requiring less than 15 minutes of human work without intervention 86% of the time.
medium technology
OpenAI's data indicates that models are not close to fully autonomous recursive self-improvement.
medium technology
Anthropic reported that internal AI models collaborated on or led more than 90% of R&D tasks, while autonomously completing zero AI R&D tasks.

Comments

Login to view and post comments.

🏠Portal 📰Links ❓Q&A 📅Events 💼Jobs