Companion to Contra Acemoglu On AI (ACX-154, 69), and a strong AI-explainer in its own right. Scott dismantles common AI-skeptic moves: the self-awareness-is-spooky riff (Dijkstra's 'does a submarine swim' — the sub can nuke you anyway), the narrow-vs-general-AI-is-a-spook argument, and the key 'blob of learning ability' framing (AlphaZero generalizing from Go to chess/shogi; DeepMind's open-ended agents; the human/rat/insect-brain scaling intuition) for why steady generalization, not a magic self-awareness module, is the path to AGI. He engages the motte-and-bailey of AI-risk arguments, gives his own estimates (75% vs Metaculus's lower numbers), defends AGI timelines ('it's the #$@&ing prior'), and caps it with a metaphor-fest (Constantinople/gunpowder, the fifty-foot dog, Neolithic-man/Burj-Dubai) plus the sharp 'only worry about the most medium-term risks' satire.

Quality 71 · Strong. Strong (71, upper companion band). Contains some of Scott's clearest AI-risk exposition (the blob-of-learning framing, the medium-term-risk satire); nudged just under the top companions because the parent essay was mid-tier and this remains a comment-response post.

Claude’s paradigm shift 52 · Moderate. Moderate (52). The blob-of-learning and generalization framings are lucid syntheses of existing views, not a new frame.

Real-world impact 1 · Negligible. 1 — a companion commentary; negligible independent real-world effect.

Highlights From The Comments On Acemoglu And AI