:( tough times...

oooof.. anyone know of any successful appeals of a submitted NIH proposal being "administratively withdrawn"?

@spiantado.bsky.social ;)

by Robert Gonsalves

Indeed! sapir.psych.wisc.edu/papers/lupya...

The relatively modest implications of multimodality for reasoning are a "dog that didn't bark" in the last 6-7 yrs. We're starting to forget now, but it seemed intuitive to many observers 1970-2020 that symbols needed sensory grounding and intelligence would need embodied interaction with the world.

I should clarify that by “does have that” I mean it seems like the most parsimonious way to explain results like this. I’m not sure what the alternative is that doesn’t do violence to the idea of a causal model.

A lot hinges on the meaning of “know” :) If it means having a causal model such that tweaking representations of emotions affects behavior in coherent and predictable ways, then it does have that. If it’s the qualia of an emotional state, then no (but what possible evidence *could* tell us this?) 😳

It’s not surprising for anyone who has already been convinced by the data that LLMs have learned coherent internal models of the human umwelt. For anyone who still thinks the models are just learning a shallow representation of language, these results (the sensible steering) should be surprising.

That’s right :) I would flip it around though: the biggest advance in AI came from teaching silicon language which l, evidently, results in recreating much of human cognition.

People celebrating (unverified) reports of data centres being hit with with missiles, on Bluesky, which helpfully only runs on a few old Pentium 4 boxes in someone's garage and not in, say, data centres. Some of you people really don't understand a single thing, do you.

What if we just went back to reading a bunch of books and thinking about them together, and that was the class? With syllabi no longer than three pages.

Keep this in mind when you dismiss AI's "hallucinations". Left to our own devices *we* are not especially good at fact checking. Hence the importance of procedures and institutions.

One takeaway from my brief LLM honeymoon phase that I'll carry with me:

It made me realize just how densely & subtly compressed "reassurance-seeking" is in my language (and probably that of users without OCD too). A prompt is just SUCH a rich signal of what framings/attunements it wants validated.

he‘s not wrong… 🤷‍♂️sapir.psych.wisc.edu/papers/lupyan_bergen_201...

Omfg.. this and the Reagan Mastermind clip - just 🌟https://youtu.be/b5wfPlgKFh8?si=4ecCJfnQEoUARJiU

What a fun conversation! @mcxfrank.bsky.social @kensycoop.bsky.social !

Ooh, very cool! There's prior work on markers of valence in diverse animal sounds, right? Do you think valence is part of it?

“I do believe I will be having the honor of taking Cuba.” “Taking Cuba. I mean, whether I free it, take it. I think I can do anything I want with it,” he said. “They’re a very weakened nation right now.”

I don’t even wait. And when you’re a star, they let you do it. You can do anything. Bush: Whatever you want. Trump: Grab ’em by the pussy. You can do anything.

Trump Doctrine, ladies and gentlemen.

My expectation is that if one is communicating their opinions as a scientist, those opinions should be updated given new evidence. If, however, one has transitioned to punditry, then make it clear that one's scientific credentials are being put aside.

Not only that, but when folks repeatedly made this point to the authors, the authors responded by blocking those folks on social media. Not once have I seen an admission of having made a mistake or at least updating their conclusions in light of new evidence. 🤷‍♂️

Very happy that this paper from our lab is now out in @pnas.org! What happens when the *same* person experiences the *same* information with a *different* interpretation? Nearly the whole 🧠—well, at least nearly all association cortex—changes how it represents that information! tinyurl.com/p8chj2j7

Speaking of Claude [Shannon], an apt quote from the man himself: "I’m a machine and you’re a machine, and we both think, don’t we?" :)

Will read as soon as I have some bandwidth!

Ah ok, it sounded like you provided the raw data to chatgpt

Large Language Models as General Pattern Machines We observe that pre-trained large language models (LLMs) are capable of autoregressively completing complex token sequences -- from arbitrary ones procedurally generated by probabilistic context-free ...

I apologize if I’m misunderstanding the context, but I think you may be misunderstanding what transformers do. They’re not “invention” machines, they’re pattern detectors. It’s analyzing the data, albeit not equivalently to a simple regression arxiv.org/abs/2307.04721

They are

Very timely! Just to clarify - Westwood claims his LLM bot passed reverse shibboleths. Are you saying that bots/agents that’re actually being deployed on crowdworking sites tend to fail these?

Of all the analogies, this one about horses is the dumbest.

📣📣📣

Applications for the 2026 Diverse Intelligences Summer Institute (DISI) are now open!!

Are you interested in intelligence, mind, and cognition in all its forms? Early-career scholars from any discipline—and storytellers in any medium—are encouraged to apply!

More info: disi.org