Lightnews — Scholar-powered news

Reposted by Jake Quilty-Dunn

Melanie Mitchell @melaniemitchell.bsky.social · 1d

Do AI reasoning models abstract and reason like humans?

New paper on this from my group:

arxiv.org/abs/2510.02125

🧵 1/10

Do AI Models Perform Human-like Abstract Reasoning Across Modalities?

OpenAI's o3-preview reasoning model exceeded human accuracy on the ARC-AGI benchmark, but does that mean state-of-the-art models recognize and reason with the abstractions that the task creators inten...

arxiv.org

3 22 79

Jake Quilty-Dunn @quiltydunn.bsky.social · 2d

i think victor gomes had the idea to write a dramatic novel about the goings on of John, Mary, Ann, and Bill in linguistics papers as one continuous narrative

1 3

Jake Quilty-Dunn @quiltydunn.bsky.social · 3d

super clear and helpful commentary by ned, highly recommend

Ian Phillips @ianbphillips.bsky.social · 3d

@neddo.bsky.social's very nice commentary on @matthiasmichel.bsky.social and @smfleming.bsky.social's BBS target article also arguing that conscious perception may form fast even if postdiction suggests it only "vulcanizes" slowly.

Ned Block @neddo.bsky.social · 3d

Hmmm…my reply is somewhat similar drive.google.com/file/d/1cIUk...

4

Reposted by Jake Quilty-Dunn

Merriam-Webster @merriam-webster.com · 7d

We are thrilled to announce that our NEW Large Language Model will be released on 11.18.25.

650 8.2K 25K

Jake Quilty-Dunn @quiltydunn.bsky.social · 6d

"nonsense" is a bit harsh!

1 5

Reposted by Jake Quilty-Dunn

Dorsa Amir @dorsaamir.bsky.social · 9d

While this looks like it could have been painted yesterday, it’s actually a 1,700 year old (!) portrait from a Fayum mummy in modern day Egypt. This is one of ~900 of these portraits from the era, which broke from a more stylized tradition, and represented the subject more naturally.

A naturalistic portrait of a woman’s face from a Fayum mummy

1 4 27

Jake Quilty-Dunn @quiltydunn.bsky.social · 13d

thanks so much elliot!

1

Jake Quilty-Dunn @quiltydunn.bsky.social · 13d

it's beautiful work, and it's exactly the sort of thing we had in mind as an alternative to model-based planning. rewarding randomly generated activity to build a route in a spacetime attractor network is v different from using an internal model to simulate counterfactuals and select among options

1 1 4

Jake Quilty-Dunn @quiltydunn.bsky.social · 14d

nyc baby, greatest city in the world

11

Jake Quilty-Dunn @quiltydunn.bsky.social · 15d

i don't know if there is a unified idea corresponding to that label, but i do think the sense in which perceptual processes infer distal stimuli from proximal stimuli involves very different computational mechanisms from the ones people use to infer q from p and if p then q

1 1

Jake Quilty-Dunn @quiltydunn.bsky.social · 15d

there are a few nonverbal questions here, including:

does the same sort of process that occurs consciously occur implicitly, just without awareness?

are the implicit processes constrained by the same principles (e.g. physical realizability) as deliberation?

we think the answer to both is no

1 2

Reposted by Jake Quilty-Dunn

johnwkrakauer @johnwkrakauer.bsky.social · 15d

First shot across the bow from ongoing project with Jake.

Jake Quilty-Dunn @quiltydunn.bsky.social · 16d

New publication forthcoming in BBS, co-authored with John Krakauer: a commentary on @smfleming.bsky.social & @matthiasmichel.bsky.social's groundbreaking target article.

We critique widespread assumptions in cognitive neuroscience about the role of internal models in implicit cognition. (1/7)

3 5 16

Jake Quilty-Dunn @quiltydunn.bsky.social · 16d

Finally, thanks to Dan McNamee, whose work on hippocampal microdynamics was an inspiration to us, for conversations and feedback.

Here is a link to a preprint of our BBS commentary:
https://philpapers.org/rec/KRAPCP-2 (7/7)

1 7

Jake Quilty-Dunn @quiltydunn.bsky.social · 16d

This is part of a larger project John and I are working on about the mysterious nature of deliberative thought and why we should resist attempts to reduce it to implicit, offline processes. (6/7)

1 1 6

Jake Quilty-Dunn @quiltydunn.bsky.social · 16d

What about explicit, deliberative weighing of alternatives? We argue that this capacity is rare in the animal kingdom and thus cannot be necessary for perceptual consciousness. We conclude that the evolved function of perceptual consciousness is unrelated to model-based planning. (5/7)

1 1 5

Jake Quilty-Dunn @quiltydunn.bsky.social · 16d

There is no need to simulate multiple alternatives according to an internal model and then select one in order to navigate the environment. F&M articulate their view of perceptual consciousness in terms of this kind of implicit model-based planning, which we think does not exist. (4/7)

1 2 3

Jake Quilty-Dunn @quiltydunn.bsky.social · 16d

We argue that the data do not support this (surprisingly common) interpretation. Hippocampal replay is often spatially fractured and physically unrealizable, and lacks the small-scale dynamics that would be necessary to support model-based planning. Other functions are more plausible. (3/7)

Figure 2 from McNamee, D. C. (2024). The generative neural microdynamics of cognitive processing. Current Opinion in Neurobiology, 85, 102855. The figure describes multiple functions of hippocampal replay without positing model-based simulation.

1 1 4

Jake Quilty-Dunn @quiltydunn.bsky.social · 16d

Many researchers think that cognitive maps in the hippocampus are used for implicit simulations in humans and non-human animals, both during waking states and sleep. This view is based on activation patterns that correspond to spatial information and aid future navigation. (2/7)

Figure 2 from Carr, M. F., Jadhav, S. P., & Frank, L. M. (2011). Hippocampal replay in the awake state: a potential substrate for memory consolidation and retrieval. Nature neuroscience, 14(2), 147-153. The figure shows rodent hippocampal sharp wave ripples during navigation in awake replay, both local and remote.

1 1 4

Jake Quilty-Dunn @quiltydunn.bsky.social · 16d

New publication forthcoming in BBS, co-authored with John Krakauer: a commentary on @smfleming.bsky.social & @matthiasmichel.bsky.social's groundbreaking target article.

We critique widespread assumptions in cognitive neuroscience about the role of internal models in implicit cognition. (1/7)

2 14 49

Reposted by Jake Quilty-Dunn

Nancy Kanwisher @nancykanwisher.bsky.social · Sep 5

‪@benhayden.bsky.social‬
@tyrellturing.bsky.social‬
@jmgrohneuro.bsky.social‬
@pessoabrain.bsky.social
I see a lot of talk on here about how we should avoid
"x does y" talk because the brain is "a dynamic, reverberant, reciprocally interconnected system".
But this does not follow.
A thread...

4 27 85

Reposted by Jake Quilty-Dunn

Evan Westra @evanwestra.bsky.social · Aug 31

August 31st/September 1st on leave

Homer Simpson in a hammock wearing a Hawaiian shirt and straw hat pouring beer into a coconut

1 1 18

Reposted by Jake Quilty-Dunn

Yoav Goldberg @yoavgo.bsky.social · Aug 27

When reading AI reasoning text (aka CoT), we (humans) form a narrative about the underlying computation process, which we take as a transparent explanation of model behavior. But what if our narratives are wrong? We measure that and find it usually is.

Now on arXiv: arxiv.org/abs/2508.16599

Humans Perceive Wrong Narratives from AI Reasoning Texts

A new generation of AI models generates step-by-step reasoning text before producing an answer. This text appears to offer a human-readable window into their computation process, and is increasingly r...

arxiv.org

4 22 85

Jake Quilty-Dunn @quiltydunn.bsky.social · Aug 27

imagine a world without LLMs. what would be the point of living

Ars Technica @arstechnica.com · Aug 27

AI advocates have warned that if every author in the class action filed a claim, it would "financially ruin" the entire industry.

Authors celebrate “historic” settlement coming soon in Anthropic class action

Advocates fear such settlements will “financially ruin” the AI industry.

arstechnica.com

8

Reposted by Jake Quilty-Dunn

Rafe Meager (they/them) @economeager.bsky.social · Aug 24

I love jargon, i think it's often both very useful and very beautiful language, and i have a crazy working theory that if you just stop having the reaction of a whiny baby when confronted with something you don't immediately understand then you will learn to love and enjoy jargon too

9 10 120

Reposted by Jake Quilty-Dunn

David Papineau @davidpapineau.bsky.social · Aug 20

We’re hiring—one year post in Ethics—urgent—deadline this time next week www.kcl.ac.uk/jobs/122985-...

Lecturer in Ethics Education (AEP)

www.kcl.ac.uk

1 24 29