『Ginevra Davis — The Attractor States of Intelligence』のカバーアート

Ginevra Davis — The Attractor States of Intelligence

Ginevra Davis — The Attractor States of Intelligence

無料で聴く

ポッドキャストの詳細を見る
Ginevra Lily Davis, the contemporary philosopher Erik calls his favorite, joins the show to defend an unusual position with disarming clarity: that the universe has a real bottom, and that bottom is positively-valenced consciousness. From there she takes on postmodernism (it eats itself), Eliezer Yudkowsky (he is the paperclip maximizer in his own thought experiment), and the deepest fear of the AI doom community (any sufficiently intelligent system will pass through truth-seeking, discover the same bottom, and arrive at the same good attractor state we did). What surprises Erik and Guy is how cleanly Ginevra's metaethics handles the practical question Erik brings near the end: he is encoding "promote the good, the true, and the beautiful, expand consciousness, reduce suffering" as a meta-filter on the agents he is building, and he wants to know if that is metaphysically sound. Her answer: yes, and more sound than Anthropic's Claude constitution.Key Topics[00:00 - 02:50] Cold open — Erik returns from a ski week — Erik's agent-pilling story: built "AI Class for Seniors" and a scam-checker for his mom in a week using a back-ordered cloud Mac Mini, after Josh Lehman's episode two weeks prior[02:50 - 13:25] What metaethics actually is — Ginevra catches Erik "smuggling" normative claims into his evolved-cooperation theory of ethics; the two-level frame (ethics-as-law-and-norm vs the brass-tacks question of why the project is necessary at all)[13:25 - 25:00] Postmodernism, explained and disposed of — Peter Thiel's "The Straussian Moment," 9/11 as the postmodern crisis, the chocolate-as-taste vs murder-as-truth distinction; Ginevra outs herself as a hedonic utilitarian[25:00 - 31:30] Irreducibility — the medieval-torturer thought experiment against the sex/positive-experience contrast; positively and negatively valenced consciousness as the irreducible bottom; Guy connects to Plato and the good, the true, and the beautiful[31:30 - 39:30] Superintelligence — what Yudkowsky-style recursively-self-improving singletons predict; why current LLMs feel more "mushy biological swarmy" than that; value attractor states; "intelligence passes through truth-seeking"; the Joe Carlsmith line that Yudkowsky himself is the paperclipper in his own thought experiment[39:30 - 41:40] Closing radio segment — plug for Arena Magazine's silicon coffee-table book due online in a week or two; Ginevra contributed the meaning-of-life-is-to-mine-the-silicon chapter--- [Act 2: Extended Conversation] ---[41:40 - 56:30] Does superintelligence need to be conscious? — Ginevra's answer: no, but it would have systematic prediction errors with a consciousness-free world model; value vessels vs value stewards (with Mike Johnson); Erik introduces Peter Watts' Blindsight and the zombie-intelligence thesis; Nick Land's popular reading vs what he told Ginevra in person ("a superintelligence wouldn't be conscious, that's ridiculous")[56:30 - 1:08:00] Why nobody wants 1984 — suffering is energetically expensive; Scott Aaronson's "Why I'm Not Terrified of AI"; bad ideas get out-competed by good ones; explicit Plato and the good/true/beautiful; Mike Johnson's symmetry theory of valence connected to Platonism[1:08:00 - 1:17:30] Postmodernism as psychic shock — Dada, WWII; the asymmetric claim that marriage isn't metaphysically grounded does not mean nothing is; Guy's logos-as-gathering thread; atemporal value as "area under the curve" between Big Bang and heat death[1:17:30 - 1:32:30] Erik's practical question and the close — Erik is encoding good/true/beautiful + expand-consciousness + reduce-suffering as a meta-filter on his agents; Ginevra: "more metaphysically sound than Claude's constitution"; Amanda Askell and the virtue-ethics framing of Claude's constitution; steel-manning anti-hedonic-utilitarianism (the "interchangeable consciousness dust" worry, level discipline)Guest BioGinevra Lily Davis is a writer and metaethicist working on the foundations of morality and the implications of superintelligence. She contributes to Arena Magazine and collaborates closely with Michael Edward Johnson on the symmetry theory of valence and related work. Previously known for her social-theory writing, her first major AI-focused essay, on the thesis that the meaning of human life is "to mine the silicon," appears in Arena's forthcoming silicon coffee-table book.Notable Moments[~07:20] The frame. — "I would say I take the possibility that goodness and badness are just human illusions and its implications more seriously than a lot of people."[~17:30] Postmodernism eats itself. — "Postmodernism has a hard time defending itself. It doesn't provide a 'why be right' about this reality in which there is nothing to be right and wrong about. It sort of eats itself."[~24:00] The medieval torturer. — "What is a torturer doing? They are trying to add maximal negativity. You can't really describe what is going on in those moments without invoking ...
adbl_web_anon_alc_button_suppression_t1
まだレビューはありません