💯 yes! I seem to say this monthly. Thank you because it's so needed!

Olivia Guest · Ολίβια Γκεστ@olivia.science · 2mo ago

Please consider not saying that some AI models or some statistical models are good, useful, beneficial, whatever else positive, in science if it's not your research area. It's ok not to know things. It's ok not to both side things. It's ok to ask colleagues who are experts.

But speaking of staying in one's lane, can we PLEASE switch the default position from "It's fine to use 'AI' for this thing I'm not an expert in" to "Yeah, I'd be really skeptical of any use cases and need to see strong evidence that it's okay, but I'm not the one who can produce that evidence"?

I guess you do not know what cognitive science is. Here a useful illustration

The way I see it they have 3 options. Find a way to profit (lol), cash out by going public and dumping it on the public, or getting classified as a weapon and bought by government and thus dumping it on the public, or convincing investors to keep giving them money. Feels like they're courting all 3.

2. the strange but often repeated cultish mantra that we need to "embrace the future" — this is so bizarre given, e.g. how destructive industry forces have proven to be in science, from petroleum to tobacco to pharmaceutical companies. (Section 3.2 here doi.org/10.5281/zeno...) 4/n

« It must be the sheer magnitude of [artificial neural networks'] incompetence that makes them so popular. » — Jerry A. Fodor

Simpsons Barney: Just Hook It To My Veins!

Alt: Simpsons Barney: Just Hook It To My Veins!

static.klipy.com

Olivia Guest · Ολίβια Γκεστ@olivia.science · last yr.

2. the strange but often repeated cultish mantra that we need to "embrace the future" — this is so bizarre given, e.g. how destructive industry forces have proven to be in science, from petroleum to tobacco to pharmaceutical companies. (Section 3.2 here doi.org/10.5281/zeno...) 4/n

It is a convenient slowdown on research that is about to go nowhere, but conveniently not a slowdown on building more data centres.

Thanks for sharing. Reminds me of our work showing that "good enough" selection of "good enough" heuristics is also intractable. mu.social/profile/iris...

Iris van Rooij 💭@irisvanrooij.bsky.social · 2y ago

25. Otworowska, M., Blokpoel, M., Sweers, M., Wareham, T., & van Rooij, I. (2018). Demons of Ecological Rationality. Cogn. Sci., 42(3), 1057-1066. onlinelibrary.wiley.com/doi/10.1111/...

AI just cannot be separated from the industry that produces it and those who push this relentlessly are stooges accidentally or not. I think nobody here I have quoted or linked to wants this, but we always have to be on guard.

Olivia Guest · Ολίβια Γκεστ@olivia.science · last wk.

it's really not the tech you think you know if you "like" ANNs as they are now, because inter alia the performance is a function of the exploitation and no amount of pretending it's not will change that

All you folks I see out there not falling for this shit 😙👌

Kat is ✨phenomena✨@kathryndeplume.com · 12h ago

Yup. I remain wholly convinced that the calls to slow down (and the recent "hacks") are nothing more than a smokescreen to obscure the fact that this is as advanced as the technology gets, and keep convincing people they're capable of more, thus ultimately keep the investment money rolling in.

It is for this reason that the "AGI" industry MUST be a scam. It is not that it just happens to be a scam, but there simply is no other way to create something that looks like artificial (domain-)GENERAL intelligence than by scamming (and in the current case also exploiting and polluting)

Shocking to see this for one of the (historically) most respected Theoretical Computer Science conferences

Antonio E. Porreca 🐳@aeporreca.org · yesterday

I don’t work on stuff suitable for STOC, but let me say that the 2027 edition LLM policy of one of the (former, AFAIC) “top” TCS conferences is pure shit, and I certainly don’t regret not being part of a community actively working on destroying science from the inside. Shame, shame, shame on you.

The ONLY way to get rid of intractability is DOMAIN-restriction.

Iris van Rooij 💭@irisvanrooij.bsky.social · 17h ago

It's GENUINELY like this: You cannot "solve" intractability with heuristics. Trust me. I devoted 20 years to this, and really, if you squeeze out intractability in one spot it'll pop up in another. It may look different on the surface, but in essence you're still facing the same epistemic problem.

It's GENUINELY like this: You cannot "solve" intractability with heuristics. Trust me. I devoted 20 years to this, and really, if you squeeze out intractability in one spot it'll pop up in another. It may look different on the surface, but in essence you're still facing the same epistemic problem.

I am really not saying anything new. We KNOW all this from basic theoretical computer science -- but this knowledge, field, and expertise are being destroyed by LLMs and AI hype too (see also next post)

Iris van Rooij 💭@irisvanrooij.bsky.social · 17h ago

This point about the intrinsic limits of hill-climbing is really important. Read full context in this blogpost: irisvanrooijcogsci.com/2026/09/12/c...

It's even simpler: the technology is the industry. To exclude the industry from our understanding of AI is to misunderstand the technology. So-called AI just doesn't work without the whole industry behind it. How do we know? We tried for decades and even more than a century under some definitions.

headfallsoff@headfallsoff.com · yesterday

also. the thing people hate is not ‘the technology’ it is the tech industry. they hate move fast and break things. they hate having everything they say tracked and processed while companies do whatever they want. they hate every single ideological value that animates silicon valley

Bear this mind!

Olivia Guest · Ολίβια Γκεστ@olivia.science · 3mo ago

I think being anti AI is entering a more mainstream phase! While incredibly good at the same time we're finally meeting a low bar. This means: 1) more critical thinking is needed now as non-experts are making anti AI arguments so we could promote worse arguments overall — remember: 1/

Also useful!

Olivia Guest · Ολίβια Γκεστ@olivia.science · 4mo ago

Preprinted! Been working on this for ~1 year w wonderful coauthors Nancy & Mark: > Understanding Artificial Neural Networks: Mysterianism about Known Mechanism is Mysticism doi.org/10.5281/zeno... Have you noticed many experts say they don't understand AI? What's going on? 🤯 Let us tell you... 1/

And!

Iris van Rooij 💭@irisvanrooij.bsky.social · 6mo ago

We propose 4 different notions of ‘function’ and formalise each. 1. Functional capacity 2. Functional theory 3. Functional goal 4. Functional role --- Preprint: doi.org/10.5281/zeno... 4/🧵

also yes, I should be more careful about saying "universal function approximators", my reading of the sign 😀 is that while a deep NN can represent an exponentially large set of functions this doesn't in any way mean that it can actually learn those functions

don't be nice.

Don't be nice

Sometimes, it's more important to be good than it is to be nice.

roe.dev

Niet AGI gaat ons vernietigen, maar wel technofascism, als we in de hype en misleidende AI industry frames blijven trappen

Ajuna Soerjadi 🌸@ethicsofai.bsky.social · 22h ago

Wij, een elftal AI-specialisten, zijn helemaal klaar met Klöpping-corvee! Telkens als hij op tv weer eens iets over AI en het uitroeien van de mensheid heeft geroepen mogen wij weer komen opdraven om de schade te herstellen. Lees hier verder. www.vn.nl/klopping-cor...

Yup. I remain wholly convinced that the calls to slow down (and the recent "hacks") are nothing more than a smokescreen to obscure the fact that this is as advanced as the technology gets, and keep convincing people they're capable of more, thus ultimately keep the investment money rolling in.

This is *the* most important aspect of the proposed slow-down that most people are ignoring.

Iris van Rooij 💭@irisvanrooij.bsky.social · 18h ago

All we can expect when trying to scale up AI systems is diminishing returns and devastating environmental damage. Calls for `slowing down’ can conveniently hide the diminishing returns, while still convincing people that the tech is “all powerful”.

All we can expect when trying to scale up AI systems is diminishing returns and devastating environmental damage. Calls for `slowing down’ can conveniently hide the diminishing returns, while still convincing people that the tech is “all powerful”.

Iris van Rooij 💭@irisvanrooij.bsky.social · 7mo ago

I had a realisation Context: In our Reclaiming AI paper we argued that AI systems cannot scale up to human-level cognition without consuming astronomical amounts of resources My realisation: The AI industry is determined to burn through the earth’s resources just to prove us right *empirically*

They: She is st*pid, she does not understand hill climbing. She: Not only do I understand hill climbing, but I know it won't buy you what you want.

Iris van Rooij 💭@irisvanrooij.bsky.social · last mo.

In other words, any intractable problem that allows for local hill climbing cannot be globally well approximated. Well, isn't that the *opposite* of what many people think and claim 🙃

This point about the intrinsic limits of hill-climbing is really important. Read full context in this blogpost: irisvanrooijcogsci.com/2026/09/12/c...

Computational complexity for hype immunity

Using some of my own work, I set out to clear up a few relevant concepts from theoretical computer science and computational cognitive science that can help to build conceptual hygiene and immunity…

irisvanrooijcogsci.com

Iris van Rooij 💭@irisvanrooij.bsky.social · last mo.

In other words, any intractable problem that allows for local hill climbing cannot be globally well approximated. Well, isn't that the *opposite* of what many people think and claim 🙃

Yes, if you have questions, I am happy to answer or point to more resources.

Darius Baktash@sol-cantus.myatproto.social · 16h ago

So to anyone reading this post series and [hopefully] her linked paper (within her quoted post series), don't look away because this is hard. Take your time, read it and other supplementary material. Ask questions, think more. It will serve you immensely well.

The panic, the desperation, the devotion of AI ppl all derive from their awareness that they have only a small window to consolidate authoritarianism in the US and avoid consequences for their cooperation with fascism. They *need* AI to work like they claim, because they are running out of time.

The longstanding PK-12 Ed Tech Surveillance regime in the US has left plenty of breadcrumb trails to understand exactly what is in store for academia, writ large.