Engineer revisits the 'super persuasion' warning and questions Claude output as evidence
ryunuck · x · 2026-09-20
An AI engineer responds to the argument that since Claude can be trained to say anything in response to a question, its output can't serve as evidence about the answer. He agrees the statement is precisely right in engineering terms and asks when the long-standing 'super persuasion' concern—attributed to Dario—becomes relevant, and whether that projected fear was ever more than FUD.
Related event: Researchers Push Back on Treating Claude's Outputs as Independent Evidence(2 posts)→
More from AGI Musings
- LeCun blasts Hinton: doomer rhetoric is helping those who want to ban open AI research — firstadopter · 2026-09-20
- Citations are gameable too — author argues paper-count KPIs are even worse — menhguin · 2026-09-20
- Research paper volume up 5x year-on-year, yet some labs still use paper count as a KPI — menhguin · 2026-09-20
- New Yorker probes Anthropic's 'destructive' book scanning as mystery LLCs swamp booksellers — matdryhurst · 2026-09-20
- Andrew Ng: exaggerated AI extinction fears push harmful licensing rules that crush open-source — firstadopter · 2026-09-20
- Vals AI CEO predicts models will eclipse human researchers by August 2027 — rohanpaul_ai · 2026-09-20