Pavel @spavel.bsky.social · 23/02/2026Hallucinations compound: 1) A worker uses an LLM to generate a report. 2) A manager uses an LLM to extract insights from the report. 3) An exec uses an LLM to turn the insights into a memo. 4) Another manager uses an LLM to summarize the memo. Each person human-washes the text for the next person. 1317538
Larry Cooper 🇨🇦 @cooperlk99.bsky.social · 23/02/20261)staff document a detailed report on a problem that is an existential threat 2) a manager asks for slimmed down version so they can understand it 3) a director asks for slide deck of 5 slides or less 4) a VP asks for a single 24-point slide for the exec team . ...management since...? 130
Alberto GM @beto.betoworks.com · 23/02/2026In AI we've been sold "the future of work"; we got its mega turbo enshittifier agent instead. This is fine 🔥☕ 000
Rachel Happe @rhappe.bsky.social · 23/02/2026An abstraction of a probability of an abstraction…media.tenor.coma cartoon of a green snake with potato-yi written on the bottomALT: a cartoon of a green snake with potato-yi written on the bottom 010
Pavel @spavel.bsky.social · 23/02/2026By the time we get to step 5 (people having to actually do some work), they might have questions for the manager, but the manager doesn't have any answers that aren't already contained in the summary, because they didn't read the memo (but even if they had, it wouldn't have helped). 1612
Patrick M. Hausen @sweordbora.hausen.com · 23/02/2026Anthony Moser@anthonymoser.comI'm imagining an instructor somewhere making a syllabus with chat gpt, assigning reading from books that don't exist But the students don't notice, because they are asking chat gpt to summarize the book or write the essay 060
Johannes Klingebiel @klingebeil.bsky.social · 23/02/2026Lindsay deWit Pratt had a pretty good way of illustrating the risk with even more benign tooling bsky.app/profile/klin...Johannes Klingebiel@klingebeil.bsky.socialThis is a pretty smart way of illustrating the cascading risks of automating research with LLMs and generative systems. www.lindseydewittprat.com/research-ris... 110
SD @trisaratops.bsky.social · 23/02/2026Love to receive a product document only to discover that it was generated by AI. The purpose is shared understanding! What would I gain by reading something you didn't write?! Why would I design based on this? 😭 030
Correlated Noise @correlatednoise.net · 23/02/2026I've experienced this on email chains I've been CC'd into and realized, upon reading from the beginning, not a single person has read any other person's email or the initial boilerplate status update. Like an oroborous, new person freaks out, adds someone else, they freak out, adds someone else etc 180
Giskard R. Reventlov @giskard.bsky.social · 23/02/2026In (sadly a sex pest) Asimov's "The Machine That Won the War" humanity feeds data into a supercomputer that then tells them were to send the fleet in their existential space war. The guy gathering the statistics suspects they are unreliable, so he makes the numbers more realistic (his guess). Then 140