Superintelligence: No one is smart enough
summary
Claims measurable results: GPT‑4 hits IQ 155 on WAIS‑3, scores in the 99th percentile on the Torrance creativity test and 99.5th on the USA Biology Olympiad, and achieves 1410/1600 on the 2022 SAT; models run with a ~24,000‑word context window (Claude 75k), train on tens of millions of books (roughly 40M cited) and tens of terabytes of data, power vision and code tools (GPT Vision, Advanced Data Analysis), pass professional exams (USMLE, bar, CPA) and are being embedded in humanoid robots (1X Neo, H1, Figure‑01).
AI-written summary · about
chapters (25)
- 0:00Introduction
- 1:10Stop talking about ChatGPT
- 3:38The pioneers of artificial intelligence
- 5:42Human intelligence extremes
- 7:29Transformers and predictive text
- 9:12Data scale in LLMs
- 11:17GPT-4 vs ChatGPT
- 12:22GPT-4 IQ benchmarks
- 15:12GPT-4 biology exams
- 17:25GPT-4 SAT scores
- 19:53Terence Tao: smartest human
- 22:48Superintelligence approaching
- 24:36Human vs AI traits
- 25:40GPT-4 professional exams
- 26:44ChatGPT user base
- 28:34Emergent AI abilities
- 29:56AI consciousness debate
- 31:21Google DeepMind Gemini
- 32:50GPT-4 coding demo
- 36:54GPT-4 analyzing slides
- 38:44GPT-4 Vision
- 40:56Humanoid robots future
- 43:07Q&A: AI vs human
- 45:14Q&A: AI privacy
- 46:19Q&A: AI geopolitics
related talks
17mPsychology of AI
58mMirror, mirror: LLMs and the illusion of humanity
01h 05mCan you trust your (large language) model?
56mSeparating fact from fiction in a world of AI fairytales
47mThe Rise of AI: A Reality Check on Generative AI
50mKeynote: After the AI Hype – What’s Real, and What’s Next
51mOpenAI and the AI that may overtake humans
46mAI vs Humans Rewriting the Rules of Success