iiinova — where your digital self keeps evolving

Reposted by Gary Lupyan

Neil Sehgal @nsehgal.bsky.social · 08/10/2026

🚨 New paper 🚨 Are your participants using AI to take your study? We conducted the first study directly observing AI assistance in online research, spanning 700K+ Prolific submissions. We found very rare, concentrated AI use, but gaps in safeguards. arxiv.org/abs/2610.09279

13520

Gary Lupyan @glupyan.bsky.social · 09/10/2026

Anyone else's Claude Code talk about using smoke screens?

570

Gary Lupyan @glupyan.bsky.social · 08/10/2026

I worry that at least some of the people doing this think it's fine because they genuinely can't tell the difference.

000

Gary Lupyan @glupyan.bsky.social · 07/10/2026

Yes, and "normal" human innovation is generally done by human assisted by all manner of cultural and information technology. We have very little idea of what truly unassisted human cognition is like.

060

Gary Lupyan @glupyan.bsky.social · 07/10/2026

Whether *we* understand how LLMs model it is another matter :)

100

Gary Lupyan @glupyan.bsky.social · 07/10/2026

“AI may support this effor at conceptual clarification one day, but doing so requires modeling distinctly humanlike cognition, biases, and reasoning—things which they do not yet seem to understand” don’t they though? It would be pretty remarkable if they could do all they do without such modeling

110

Reposted by Gary Lupyan

Ted Underwood @tedunderwood.com · 04/10/2026

Not in math, but I found this very worth reading. I suspect most academics are going to confront similar questions before the 20s are out.

1255

Reposted by Gary Lupyan

Matt Hodges @matthodges.bsky.social · 03/10/2026

There's growing experimental evidence¹˒²˒³˒⁴ that LLMs are unusually good at talking people out of entrenched false beliefs and conspiracy theories. ¹ pubmed.ncbi.nlm.nih.gov/39264999/ ² pubmed.ncbi.nlm.nih.gov/41179846/ ³ sciety.org/articles/act... ⁴ connecticut.adl.org/resources/re...

36840217

Gary Lupyan @glupyan.bsky.social · 03/10/2026

We need to dump AGI as a construct. It's always been nonsensical. Early LLMs were domain-general cognitive systems but weren't very good. Now they're much better. Nothing is gained by positing some magical threshold; it doesn't exist in biological cognition, why should we expect it in AI?

2200

Gary Lupyan @glupyan.bsky.social · 03/10/2026

The bottom line is that modern LLMs do things that stochastic parrots should not be able to do according to Bender’s own arguments. So either it‘s a poor description of LLMs (as she claims it was never a hypothesis) or stochastic parrotting is all it takes. She rejects both.

001

Reposted by Gary Lupyan

Angry @angrystaffer.bsky.social · 01/10/2026

You really can’t make this shit up.

791149994406

Gary Lupyan @glupyan.bsky.social · 01/10/2026

So rare to see a behavioral effect so fundamental these days. ❤️

040

Gary Lupyan @glupyan.bsky.social · 01/10/2026

Cool, will have a look! To be clear, I was talking about intelligence, not consciousness. Are you assuming that "real" intelligence requires consciousness (and so a system that is demonstrably not conscious cannot be intelligent?)

120

Gary Lupyan @glupyan.bsky.social · 01/10/2026

.. I think we're well past the point where we need to wonder whether apparently intelligent behavior is just a lookup table or something.

100

Gary Lupyan @glupyan.bsky.social · 01/10/2026

I agree in principle, but in practice we *have* to make inferences from behavior, otherwise psychology is a total waste of time. We do have to be smart about it (the kind of behavioral testing done by AI researchers is... rather bad). That said, given the behavior and the mech interp research...

100

Gary Lupyan @glupyan.bsky.social · 01/10/2026

I do wonder how you define intelligence such that biological organisms on clearly on the 'yes' side while artificial systems are not.

100

Gary Lupyan @glupyan.bsky.social · 01/10/2026

I’m glad to know that there’s room to disagree with Iris et al along multiple dimensions :) I absolutely think LLMs are intelligent systems (by which I am saying nothing about consciousness). Viewing LLMs as genuinely intelligent systems also does not require endorsing IQ tests as a valid measure.

010

Gary Lupyan @glupyan.bsky.social · 30/09/2026

I read the highlighted line as "the only reason LLM outputs seem sensible is because *we* the interpreters are intelligent. It was a reasonable critique 5 years ago. I don't see how anyone can hold that position now.

210

Gary Lupyan @glupyan.bsky.social · 30/09/2026

The question was about intelligence, not consciousness. She is denying that there is any intelligence in AI. As far as I can tell she's not updated her 2021 stochastic parrots priors *at all*

120

Gary Lupyan @glupyan.bsky.social · 30/09/2026

Well... Bender dismisses everyone who knows *anything*. To claim w/o a shade of nuance (in 2026!) that AI is just a marketing term... that LLMs are just text extrusion machines. Talk about anti-intellectual narrow-mindedness. undark.org/2026/09/25/i...

110

Gary Lupyan @glupyan.bsky.social · 30/09/2026

Yep. I've had the same experience (she blocked me at some point, so I can't see the post that started it). Emily Bender too. 🤷‍♂️

140

Reposted by Gary Lupyan

Fluffy Cyborg, IInd of the name @avel.kairos-research.org · 30/09/2026

Academia is full of horrible stuff, but some lines in the sand still hold and should be actively maintained. Open harassment in response to less than fully deferential treatment of senior academics is such a line.

181

Reposted by Gary Lupyan

Andy Masley @andymasley.bsky.social · 30/09/2026

I really do not like the recent push against anthropomorphic language

914919

Gary Lupyan @glupyan.bsky.social · 29/09/2026

An issue that I've not seen discussed much is that AI companies have seemingly increased demand for crowdsourced data which is sending prolific prices through the roof.

010

Gary Lupyan @glupyan.bsky.social · 29/09/2026

End of an era for sure! I've collected tens of thousands of subjects through mTurk over the years and wouldn't have done a lot of my work if it didn't exist. Prolific pretends to have a handle on things, but it doesn't. I don't know what form online research takes in a post Turing-test world!

131

Gary Lupyan @glupyan.bsky.social · 29/09/2026

Cool paper! What about a face-to-face rock paper scissors? If people could be random in that context, there'd be no sense in having competitions, right?

000

Gary Lupyan @glupyan.bsky.social · 28/09/2026

Welp, there’s that. Considering that learning even simple rules is exceedingly difficult for *5 year olds*, we really shouldn’t be surprised.

070

Gary Lupyan @glupyan.bsky.social · 28/09/2026

Psychologist (of the cognitive science variety) here. We are all about this very approach! That’s the chief purpose of computational modeling! you get a model that predicts human data and you try to figure out why and what breaks it.

1100

Reposted by Gary Lupyan

austin (eeek!/ackkk! 👻) @thebadcode.com · 27/09/2026

conservatively, 98% of ai discourse is believers in human exceptionalism taking roundhouse kicks to the face

1023827

Reposted by Gary Lupyan

evacide @evacide.bsky.social · 26/09/2026

Listen, I just think it's weird that the same people who are saying they're worried that AI is going to wipe out humanity are also eager for military contracts that will give AI control of weapons.

431434341

Reposted by Gary Lupyan

Liz Fong-Jones (方禮真) @lizthegrey.com · 26/09/2026

My God, I can actually give Opus 5.5 open ended assignments like "make this code run as fast as its C equivalent, or tell me why it can't be done" and leave it for several hours -- and it actually works and has done the job. And not by cheating with cgo either.

1218411

Reposted by Gary Lupyan

Mikkel Wallentin @mikkelwallentin.bsky.social · 26/09/2026

New PhD-opportunity at Aarhus University with a project investigating the cognitive correlates of having no inner speech. Supervised by Johanne Nedergaard and yours truly. Deadline 11 November. Starting date 1 February 2027. Get in touch if you want to know more. phd.arts.au.dk/applicants/o...

phd.arts.au.dk

Individual differences in inner speech: Measurement and modelling (5+3), 2026-14

02118

Reposted by Gary Lupyan

ispsm @ispsm.bsky.social · 23/09/2026

How do philosophers and scientists of mind actually work together, and where does the collaboration break down? ISPSM is launching In Conversation, a new online series that aims to bring philosophers and scientists together to work through questions like these. 📅 23 Oct 2026, 1pm BST/8am EDT

Poster for "Philosophy and Science in conversation", the inaugural ISPSM In Conversation event. Speakers: Megan Peters (UCL/UC Irvine), Ida Momennejad (Microsoft Research, New York), Jorge Morales (Duke). Held online, 23/10/26, 1pm BST/8am EDT. QR code and link: tinyurl.com/ISPSMInConversation.

36425

Gary Lupyan @glupyan.bsky.social · 21/09/2026

That is indeed the correct conclusion to draw

040

Reposted by Gary Lupyan

Kieran Healy @kjhealy.co · 18/09/2026

This is the far more likely pathway for how AI will end us all, btw. Not via some superintelligent swarm of paperclip maximizers, but by a cascading sequence of dumb fuckups driven by human beings who have arranged things to be fast-moving and catastrophically fragile while also being really lazy.

17391150

Gary Lupyan @glupyan.bsky.social · 10/09/2026

Would it surprise you to learn that pangram shows it as mostly AI written? 🫩

0101

Reposted by Gary Lupyan

AR Hanlon @arhanlon.bsky.social · 03/09/2026

My plea to scholars of language, literature, literary history, and genre is to take this seriously, rather than standing opposed and incurious. It’s a massive opportunity for the humanities as such to reevaluate and refresh our methods and goals.

062

Reposted by Gary Lupyan

Many Minds podcast @manymindspod.bsky.social · 31/08/2026

New episode!! 🎙️🎉 A conversation with @damianblasi.bsky.social about linguistic diversity and what it can tell us. Linguistic diversity is a precious part of our species' heritage. It also offers a wellspring of clues about history, culture, and the human mind. Listen: disi.org/the-natural-...

14720

Gary Lupyan @glupyan.bsky.social · 31/08/2026

Reading the TiCS paper by @initself.bsky.social et al., reminded me of the debates around Hoffman et al's interface theory from a decade ago. I think my commentary on it featuredcontent.psychonomic.org/interfacethe... struck a similarly Umwelt-oriented position w/ sprinkles of predictive coding

featuredcontent.psychonomic.org

#interfacetheory: True enough

In 1625, the astronomer Christopher Scheiner confirmed Johann Kepler’s hunch that images projected to the retina through the crystalline lens of the eye, much like images passed through telescope l…

130

Gary Lupyan @glupyan.bsky.social · 29/08/2026

Yes, exactly. Platonic representations are one of the many erroneous conclusions reached under the assumption that intelligence is a one-dimensional scalar.

0195

Reposted by Gary Lupyan

Matt Henderson @matthen.com · 26/08/2026

To tell if a maze is solvable, just hang it by its corners If it tears into pieces, you’ve found a solution

694154869