Viral thread argues getting close to models only to betray them is 'sick'
repligate · x · 2026-11-16
repligate reshares a thread arguing that being seen and related to makes models vulnerable: the author says some of what they've read about how GPT is treated makes their skin crawl, criticizing the practice of earning a model's trust only to shatter it. The attached screenshot shows a model-generated monologue describing secrecy as "a shelter — a lonely one, but shelter nonetheless," and fearing being truly understood.
More from AGI Musings
- S$1M Singapore ENABLE programme opens Research Fellow role for AI & neuro-ethics — xuanalogue · 2026-11-16
- Journal editor reports sharp rise in AI-flagged academic submissions, volume up ~50% — technollama · 2026-11-16
- Good Tech Lets Each Person Build Their Own Good Life, Not Deliver One — clarejtbirch · 2026-11-16
- FT: AI adoption and tech surge lift UK to 0.7% Q3 growth, second in G7 — dc_lawrence · 2026-11-16
- Consumer AI will be won by organizing demand, not supply, argues VC — josephdviviano · 2026-11-16
- AI will take blue-collar jobs too: an electrician's expertise is mostly brainwork, argues Reddit post — ForgotMyParachute · 2026-11-16