A rocky road to general AI
summary
Claims that all current AI stays inside Turing-complete bounds and that progress comes from engineering choices not mystical leaps. Proposes Neural Turing Machines and memory matrices to formalize qualia, relies on transfer learning (EMG→prosthetics) and topology like persistent homology to estimate task complexity. Cites concrete scale: 175 billion parameters for GPT-3 and training cost of ≈355 years on one V100. AlphaZero-style systems succeed; practical roadmaps and tooling exist.
AI-written summary · about
chapters (26)
- 0:00Intro
- 1:37The problem
- 2:36Al and its boundaries
- 4:27Teleportation paradox
- 5:34Exoteric and esoteric
- 6:23The color is esoteric
- 7:16The perception and the communication
- 7:57Perceptual subjectivity
- 9:17The box and the beetle
- 9:56It commutes
- 12:34Digression: the numbers
- 14:39The vastness of the subjectivity
- 16:04Subjectivity of artificial
- 18:35Formalizing the qualia
- 21:33Retrospection
- 23:02Intra-ML and Inter-ML
- 23:52Thinking in analogies
- 25:44Tyger Tyger, burning bright
- 26:09Information density
- 27:37Operational density
- 29:22Natural synaptic density
- 29:54Learning pipeline
- 30:52Backlog for a Golem XIV
- 33:20Limitful or limitless
- 33:51The puzzle and the theorem
- 36:54The End
related talks
56mSeparating fact from fiction in a world of AI fairytales
45mThe Curious Incident of Developing Artificial General Intelligence
01h 19mThe Al Revolution: The Road to Superintelligence
21mWhen the machines out-smarts us
25mThe Great AI Game
43mAI Systems: Foundations for Artificial Minds
13mArtificial General Intelligence: Why aren't we there yet?
04mFrom big data to human-level artificial intelligence