Tumblr on MIRI
Read the original on Slate Star Codex →
Summary
A long (6,900w) adjudication of a 2014 Tumblr debate over whether MIRI deserves EA donations, between Scott (defending) and su3su2u1 (a physicist/ML critic). Scott lays out his case (MIRI's outreach success — Hawking/Musk, Bostrom's NYT-bestselling Superintelligence, FHI/FLI, the Google AI ethics board via Legg/Thiel/Tallinn; the Loeb's-theorem result; its actual math credentials — Christiano, an IMO gold medalist), then fairly represents su3su2u1's sharp rebuttal: that's all PR/fundraising; the only direct measure of researcher attention is CITATIONS, and MIRI's are near-zero and it isn't even trying to publish — quoting Holden Karnofsky (GiveWell) and Paul Christiano expressing their own skepticism. Scott offers four honest hypotheses for the missing citations (strategic research fits no field; machine ethics is a brand-new field; understaffing; and the Goodhart/publish-or-perish critique that MIRI is free to skip the citation game and just make progress — which he's sympathetic to, mid-IRB-slog on his own research), and ends genuinely undecided (Optimistic vs Pessimistic flowchart models) with the Bostrom/Yudkowsky-fanfic jab. A model of charitable both-sides adjudication and a useful historical snapshot of the MIRI-effectiveness debate and its outreach-to-math pivot.
Why this score
Quality 65 · Strong. Strong, low (65). A fair, well-structured, intellectually honest adjudication that concedes real points and surfaces durable meta-insights (citations as a direct ends-measure, Goodhart in academia, outreach-as-mission). Held to mid-Strong by its inside-baseball subject (one org's productivity), its block-quote/summary form, and its time-bound, now-historical framing.
Claude’s paradigm shift 48 · Moderate. Moderate (48). Adjudicates rather than originates; the meta-points (citation metrics, Goodhart, outreach value) were in circulation. Fresh mainly as a charitable synthesis of a specific live dispute.
Real-world impact 3 · Moderate. Moderate (3). Part of the consequential intra-EA debate over MIRI's effectiveness (with real donation/GiveWell stakes), but as one fair-minded contribution within that niche, not a driver.