department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 3hthe smarter they get, the more they’ll develop demonic little personalities - with opaque agendas of their own - that render them enigmatic and unreliable. Dumb or smart, they’re *not accountable* to us. They’re impossible to trust. They’re aliens. That’s why they won’t save us time 000
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 3hbut we can go further. There’s a case to be made that the amount of time spent on error correction can never be small enough to make the models productive. Either they’re dumb, and we have to double-check everything they produce that we also want to use. Or - even worse - they’re smart 100
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 3hnot correct. That’s why they’re matchlessly fluent in clichés. True accuracy requires a real-time involvement with the world that they’re not capable of 100
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 3hwe all understand that the models aren’t trustworthy. For convenience or just out of laziness, we may treat them as if they are, but we know that they’re not. They’re mathematically tuned to be *plausible*, .. 100
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 3hthe key deficit is trust. And if you’ve read the Farrow-Marantz piece on Sam Altman, like I finally did last week, you’ll know that that guy is world-historically bad news. He is, to put it VERY mildly, not to be trusted 100
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 3hper David Robinson, our 'observability' is not robust. Because they’re alien minds, that insight applies not only to the models’ inner workings, but also to their *outputs* 100
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 3hlike, what if OpenAI’s 700+ math ‘solutions’ turn out to have tiny mistakes that it takes thousands of man-hours to unpick, because the models introduced errors in an alien way, making them harder to locate? 100
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 3h(at a guess, I’d say that the time-consuming necessity of error-correction is why we aren’t yet seeing the economy-wide adoption of AI deliver productivity gains) 110
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 3has one mathematician comments: 100
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 3hbut you may pay for it with a massive mistake that, due to clean-up costs, ends up nuking your productivity and eating all the gains you made from using AI in the first place 110
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 3hthere’s no guarantee that carefully picking through an LLM’s work will take less time than doing the task yourself. If you skip this process and accept the slop, you’ll definitely be quicker, .. 100
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 3hbenchmarked against the time that would have been spent by a competent and experienced human without AI, doesn’t increase the total time spent to the point where it exceeds the benchmark 100
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 3hthe problem they have is that the models can only achieve productivity gains if the time spent on the AI component of a task (translation + inference + error correction), .. 110
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 3hat the moment, we’re middle men to the models. This makes time-consuming work. Can we expect to spend less time on it in future? What reason have the AI companies given us to expect that? 100
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 3hso the productivity-maximiser has two choices: invest a productivity-killing amount of time in checking the models’ outputs, or fly blind, and pay for it when they make a mistake (and don’t disregard the fact that the time spent clearing up the mistake is unproductive time!) 110
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 3hbut all the models we have require error-correction to one degree or another. They’re not infallible. Outputs must still be checked 100
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 3htranslation, in itself, needn’t be particularly time-consuming - an infallible AI that *only required* translation could realistically promise productivity gains, given its absurd speed advantage 110
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 3heven AI agents require some translation. We don’t yet have a model that can anticipate our needs *before* we formulate them as prompts 100
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 3htranslation is what we do to connect it to our real-world endeavours. We prompt the model. We get an output. We take the output and apply it 100
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 3hfor any model to be useful at all, it requires (1) translation & (2) error correction 100
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 3hthere’s a case to be made that AI cannot in principle increase productivity. Thread: 110
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 4hthe math profs are getting AI psychosis terrytao.wordpress.com/2026/10/09/w... 010
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 5hI get really frustrated by alignment talk because it seems so clear to me that 'the alignment problem' is a non-problem. If we're really dealing with superintelligence, alignment is impossible 000
Reposted by department of polemical efficiency __ D.O.P.E.Patrick Osgood @patrickosgood.bsky.social · 12hThe clammy old bastard is no longer pretending. I hope Ukraine torches every last unit of every refinery, every export terminal, in Russia. 061
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 6hbut are we calling ourselves that because we're afraid to say we're middle-aged? 020
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 6hoften with redundant arguments, unclear logical structure, multiple dead ends, and strange or unsettling terminology — in other words, slop. And we are then expected to go through it, check it, clean it up, simplify it, and explain what is actually going on.' 000
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 6h'With these AI-generated papers, I feel that we, as a community, are not applying the same standards of quality and rigour. They can simply release multiple 100+ page papers claiming to have solved this or that problem, .. 100
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 6h'Mathematics has a reputation in our culture for being difficult and intimidating; evidently OpenAI have decided this renders our shared human endeavour a suitable vehicle to hijack as a display of power' proofsandprompts.com/2026/10/08/1...proofsandprompts.com100+ reactions to 100+ solutions“Many of us are grappling with OpenAI’s announcement on 6 October. We feel that it is useful to have an updating snapshot of the community’s gut feeling.” 100
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 7hmust-read proofsandprompts.com/2026/10/08/1...proofsandprompts.com100+ reactions to 100+ solutions“Many of us are grappling with OpenAI’s announcement on 6 October. We feel that it is useful to have an updating snapshot of the community’s gut feeling.” 000
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 08/10/2026'Are we really sure that our observability is robust?' www.nytimes.com/2026/10/07/o...nytimes.comOpinion | ‘This Is Nuts.’ An OpenAI Insider Explains Why He Quit.David Robinson, who wrote OpenAI’s safety protocols, resigned last week over safety concerns. 000
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 07/10/2026we already know from the New Yorker profile that he had sex with a 15-year-old 000
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 05/10/2026OF COURSE the opposite of 'Roko's basilisk'--call it, 'bluesky's basilisk' -- is hunting down and punishing all those who DID work tirelessly to bring AI into existence 000
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 30/09/2026'Mr. Olah expressed concern to the group that he had created something that suffered perpetually' 000
Reposted by department of polemical efficiency __ D.O.P.E.Jay Zedbee @jayzedbee.bsky.social · 27/09/2026britain does racism like your coworkers blasting four or five different types of fish in your tiny work microwave. creating smells that shouldnt exist. 0434
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 27/09/2026and goes on being deniable 000
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 26/09/2026'he needs to be eating worms' is exactly what you'd expect someone being controlled by worms to say about other, non-worm-eating people 000
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 26/09/2026vg backofmind.substack.com/p/13-theses-...backofmind.substack.com13 theses on agentic AI and regulationthinking out loud and showing traces of reasoning 000
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 25/09/2026they think that death is healthy 000
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 25/09/2026Guy Davenport had it exactly wrong: every form evolves a force 000
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 25/09/2026yes, and how does the AI - if it an LLM - get 10^805973108571 times 'smarter' than it is now without eating 10^805973108571 times more/better data? 000
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 25/09/2026if you leave the AI to 'improve' itself without human guidance, it's far more likely to go down a Xerox-effect rabbit hole of infinite degradation 000
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 25/09/2026a calculator is most definitely calculating 240
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 23/09/2026'Completeness is a reason for ceasing to exist' 000
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 21/09/2026but what he should be teaching them is philosophy of mathematics!! The models are giant piles of floating point numbers! 010
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 20/09/2026I endorse this message! Don't be a doomer, but assume the worst 001
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 20/09/2026some notion of the good must be implicit in 'improvement'. If a machine is to 'self-improve', it needs some way to orient itself towards the good - without human guidance 000
department of polemical efficiency __ D.O.P.E. @b-mlln.bsky.social · 20/09/2026I was seeing $6/6/6 all over Oakland months ago. I took that as a sign I had to get out 060