By Cory Doctorow (GPG 0xBF3D9110957E5F4C) @doctorow.pluralistic.net · 18/09/2026But - by definition - it can't contain any of our future surprises, because it's just trained on the things we already know, and if we knew about something, it wouldn't be a surprise anymore. 214210
By Cory Doctorow (GPG 0xBF3D9110957E5F4C) @doctorow.pluralistic.net · 18/09/2026Image: Zhaoxing Wang, Kunpeng Wang & Yan Xu (modified) www.nature.com/articles/s41... CC BY 4.0 creativecommons.org/licenses/by/... eof/nature.comFriction injury of the central vein caused by catheter for hemodialysis: an in vitro study - Scientific ReportsScientific Reports - Friction injury of the central vein caused by catheter for hemodialysis: an in vitro study 5390
Libby @libertybellina.bsky.social · 18/09/2026All of this makes sense to me in terms of LLMs hallucinating textual answers. But could you explain in simple terms how this relates to AI agents erasing their work or roaming outside their sandboxes when given a task? That is harder to understand. 100
anf-ingdelight.bsky.social @anf-ingdelight.bsky.social · 18/09/2026LLMs are reversion-to-the-mean machines that provide a soothing sense of control/omniscience. No statistical model is a god, but LLM AI adoption is giving BIG opiate of the masses energy. As if Big Tech accidentally reverse engineered religion as a means of control but got high on their own supply 140
Martin Austwick @martinaustwick.bsky.social · 18/09/2026This reminds me a bit of Claude Shannon’s measure of “surprisal” - that went into his informational definition of entropy 061
bigbenchrob 🍁🇨🇦 @bigbenchrob.bsky.social · 18/09/2026A useful approach to the limitations of LLMs trained on what we already know. That the DOGE punks confidently gutted science because "AI will cure cancer" incenses me. Science advances precisely at the margin between what we know and what we don't -- any real advance depends on a *surprise*. 58511
Horrificallie @grissallia.bsky.social · 18/09/2026It's one of the reasons I'm not worried that AI will "steal my job". One of the parts of my jobs is identifying and replicating edge cases. In my context, the edge cases are occurring at the rough points, and when humans interact with tech in a way that just never occurred to anyone developing it. 2222
