AI Is Now Improving Itself

In 1965, a mathematician who worked alongside Alan Turing wrote a single paragraph that has haunted AI research ever since. He predicted that one day, a machine would learn to improve itself, and that everything after that point would change. Sixty years later, that loop is starting to close. In this video, we trace how AI got here: from I.J. Good's 1965 prediction, to AlphaGo Zero teaching itself Go in 72 hours, to AlphaEvolve cracking a math problem that had stood unbeaten for 56 years, and then quietly speeding up the training of the very model that runs it. We look at the data behind the trend (autonomous AI task length is doubling every 7 months), the walls AI keeps running into (compute, data, energy), and what the people building this technology are actually saying about how close we are. This video is an honest look at what "self-improving AI" really means, and what it doesn't. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 📚 RESEARCH & SOURCES THE FOUNDATIONAL PAPER - I.J. Good (1965) — Speculations Concerning the First Ultraintelligent Machine: https://languagelog.ldc.upenn.edu/myl/Good1964.pdf REAL-WORLD SELF-IMPROVEMENT - AlphaEvolve — Google DeepMind announcement: https://deepmind.google/blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/ - AlphaEvolve — full paper (arXiv): https://arxiv.org/abs/2506.13131 - Recursive self-improvement — overview: https://en.wikipedia.org/wiki/Recursive_self-improvement THE TRENDLINE - METR — Measuring AI Ability to Complete Long Tasks (the 7-month doubling paper): https://arxiv.org/abs/2503.14499 - AI 2027 — scenario forecast by Daniel Kokotajlo et al.: https://ai-2027.com/ - Leopold Aschenbrenner — Situational Awareness: The Decade Ahead: https://situational-awareness.ai/ - Tom Davidson — What a Compute-Centric Framework Says About AI Takeoff Speeds: https://www.alignmentforum.org/posts/Gc9FGtdXhK9sCSEYu/what-a-compute-centric-framework-says-about-ai-takeoff VOICES FROM THE LABS - Dario Amodei (Anthropic CEO) — Machines of Loving Grace: https://darioamodei.com/essay/machines-of-loving-grace THE WALLS - Shumailov et al. — AI Models Collapse When Trained on Recursively Generated Data (Nature, 2024): https://www.nature.com/articles/s41586-024-07566-y - Humanity's Last Exam — Hendrycks et al.: https://arxiv.org/abs/2501.14249 WHAT EXPERTS ACTUALLY THINK - Grace et al. — Thousands of AI Authors on the Future of AI (largest survey of AI researchers ever conducted): https://arxiv.org/abs/2401.02843 - Benjamin Todd — Shrinking AGI Timelines (80,000 Hours review): https://80000hours.org/2025/03/when-do-experts-expect-agi-to-arrive/ ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ ⏱ CHAPTERS 00:00 - Intro 00:31 - We Predicted This 02:48 - Does Improvement Have A Ceiling? 04:09 - Alpha Evolve Reclusive Self-Improvement 04:45 - GPT 5.3 Codex 05:38 - Douglas Engelbart Experiment 06:41 - Counterarguments 06:52 - The ARC Test 07:51 - The Energy Problem 08:15 - Economics 08:48 - Medicine 09:20 - Is this the ceiling? 11:10 - The AI Hivemind 11:55 - Final Thoughts & Closing ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ If this video helped you understand where AI actually is, not where the headlines say it is, subscribe to our channel! Check out TAAFT: https://taaft.com/