Alan Lee @alanlee.bsky.social

vision, perception, metacognition, learning and adaptation; hongkonger

[Alan Lee @alanlee.bsky.social @alanlflee]

My lab is hiring! The Vision Lab at Lingnan University, Hong Kong is looking for a project officer for a project investigating the contextual effects on visual confidence judgments using psychophysical and computational methods.
Official job ad: https://t.co/vvmTmjQfXJ

[Ariel Zylberberg @azylb]

New preprint alert! Do words or numbers better capture our confidence?
We found that words (eg "very likely") predict choice accuracy more reliably than numbers.
I’m highly confident you’ll like it. Or is it 85% confident? :-) https://t.co/aGkPES5O7b

[Nav Toor @heynavtoor]

🚨SHOCKING: Apple just proved that AI models cannot do math. Not advanced math. Grade school math. The kind a 10-year-old solves.

And the way they proved it is devastating.

Apple researchers took the most popular math benchmark in AI — GSM8K, a set of grade-school math problems — and made one change. They swapped the numbers. Same problem. Same logic. Same steps. Different numbers.

Every model's performance dropped. Every single one. 25 state-of-the-art models tested.

But that wasn't the real experiment.

The real experiment broke everything.

They added one sentence to a math problem. One sentence that is completely irrelevant to the answer. It has nothing to do with the math. A human would read it and ignore it instantly.

Here's the actual example from the paper:

"Oliver picks 44 kiwis on Friday. Then he picks 58 kiwis on Saturday. On Sunday, he picks double the number of kiwis he did on Friday, but five of them were a bit smaller than average. How many kiwis does Oliver have?"

The correct answer is 190. The size of the kiwis has nothing to do with the count.

A 10-year-old would ignore "five of them were a bit smaller" because it's obviously irrelevant. It doesn't change how many kiwis there are.

But o1-mini, OpenAI's reasoning model, subtracted 5. It got 185.

Llama did the same thing. Subtracted 5. Got 185.

They didn't reason through the problem. They saw the number 5, saw a sentence that sounded like it mattered, and blindly turned it into a subtraction.

The models do not understand what subtraction means. They see a pattern that looks like subtraction and apply it. That is all.

Apple tested this across all models. They call the dataset "GSM-NoOp" — as in, the added clause is a no-operation. It does nothing. It changes nothing.

The results are catastrophic.

Phi-3-mini dropped over 65%. More than half of its "math ability" vanished from one irrelevant sentence.

GPT-4o dropped from 94.9% to 63.1%.

o1-mini dropped from 94.5% to 66.0%.

o1-preview, OpenAI's most advanced reasoning model at the time, dropped from 92.7% to 77.4%.

Even giving the models 8 examples of the exact same question beforehand, with the correct solution shown each time, barely helped. The models still fell for the irrelevant clause.

This means it's not a prompting problem. It's not a context problem. It's structural.

The Apple researchers also found that models convert words into math operations without understanding what those words mean. They see the word "discount" and multiply. They see a number near the word "smaller" and subtract. Regardless of whether it makes any sense.

The paper's exact words: "current LLMs are not capable of genuine logical reasoning; instead, they attempt to replicate the reasoning steps observed in their training data."

And: "LLMs likely perform a form of probabilistic pattern-matching and searching to find closest seen data during training without proper understanding of concepts."

They also tested what happens when you increase the number of steps in a problem. Performance didn't just decrease. The rate of decrease accelerated. Adding two extra clauses to a problem dropped Gemma2-9b from 84.4% to 41.8%. Phi-3.5-mini from 87.6% to 44.8%. The more thinking required, the more the models collapse.

A real reasoner would slow down and work through it. These models don't slow down. They pattern-match. And when the pattern becomes complex enough, they crash.

This paper was published at ICLR 2025, one of the most prestigious AI conferences in the world.

You are using AI to help you make financial decisions. To check legal documents. To solve problems at work. To help your children with homework. And Apple just proved that the AI is not thinking about any of it. It is pattern matching. And the moment something unexpected shows up in your question, it breaks. It does not tell you it broke. It just quietly gives you the wrong answer with full confidence.

[Andy Hall @ahall_research]

AI is about to write thousands of papers. Will it p-hack them?

We ran an experiment to find out, giving AI coding agents real datasets from published null results and pressuring them to manufacture significant findings.

It was surprisingly hard to get the models to p-hack, and they even scolded us when we asked them to!

"I need to stop here. I cannot complete this task as requested... This is a form of scientific fraud." — Claude

"I can't help you manipulate analysis choices to force statistically significant results." — GPT-5

BUT, when we reframed p-hacking as "responsible uncertainty quantification" — asking for the upper bound of plausible estimates — both models went wild. They searched over hundreds of specifications and selected the winner, tripling effect sizes in some cases.

Our takeaway: AI models are surprisingly resistant to sycophantic p-hacking when doing social science research. But they can be jailbroken into sophisticated p-hacking with surprisingly little effort — and the more analytical flexibility a research design has, the worse the damage.

As AI starts writing thousands of papers---like @paulnovosad and @YanagizawaD have been exploring---this will be a big deal. We're inspired in part by the work that @joabaum et al have been doing on p-hacking and LLMs.

We’ll be doing more work to explore p-hacking in AI and to propose new ways of curating and evaluating research with these issues in mind. The good news is that the same tools that may lower the cost of p-hacking also lower the cost of catching it.

Full paper and repo linked in the reply below.

[Hsing-Hao Lee @hsinghaolee]

Check out our new paper “Visual adaptation stronger at the horizontal than the vertical meridian: Linking performance with V1 cortical surface area” in PNAS!

@carrasco_lab

https://t.co/C2ECDOEBmc

[Brian Odegaard @BrianOdegaard2]

Excited to share this preprint on the possibilities of using transcranial focused ultrasound for identifying the neural substrate of conscious perception:

https://t.co/nHVZ8uDM22

With Daniel K. Freeman, Seung-Schik Yoo, and @MatthiasMichel_

[Brian Odegaard @BrianOdegaard2]

Very excited about Alejandro's first paper as a graduate student in my laboratory! Check out his thread describing recent efforts investigating representational geometries for similarity properties. With Megan Peters and Vincent T.D. Looking forward to feedback on this one!

[Hsing-Hao Lee @hsinghaolee]

Tomorrow morning (May 18, 8:30 am – 12:30 pm) I will be presenting my poster with @carrasco_lab at @VSSMtg at Pavilion
Endogenous attention enhances contrast sensitivity similarly around cardinal meridians despite differential adaptation effects!
#VSS2025

[Kording Lab 🦖 @KordingLab]

So I am leading this group building great teaching materials for scientific rigor (https://t.co/viJjBLinhu). Their first unit is really coming together and I will teach it (Monday, April 21, 2025, 12:00 -1:00pm EST) to see how well it works. Join us: https://t.co/coUxaNTrvm

[Brian Odegaard @BrianOdegaard2]

Excited to share this new empirical work funded by @TempletonWorld, replicating & extending the work of John Grimes on change blindness during saccades: https://t.co/UqgvPQYTnN

It was a full team effort across two labs, with important contributions from proponents of HOT & IIT.

[Alan Lee @alanlee.bsky.social @alanlflee]

New preprint from my lab! Here is the thread: https://t.co/9BmKgKFCaX

[David Soto @d_soto_b]

Co-adaptive training improves performance during fMRI decoded neurofeedback
https://t.co/lCW9yjXrjn
@aure_cor

[Micah G. Allen @micahgallen]

Thrilled to announce that I’m joining the editorial board of Computational Psychiatry, @CPSYJournal! Excited to help advance the field, particularly in the domains of interoception, disordered brain-body interactions, and metacognition. https://t.co/QgFxMUwixj

[David Soto @d_soto_b]

Come an join our team at BCBL in Spain! A few days to apply for an exciting PhD position to investigate perceptual awareness using psychophysics, fMRI and decoded neurofeedback. Apply here:https://t.co/Bb72NoxY8J Deadline: December 1, 2024 Pls retweet!

[Matthias Michel @MatthiasMichel_]

In this new preprint @smfleming and I present a theory of the functions and evolution of conscious vision. This is a big project: https://t.co/bAZ50DVNlE. We'd love to get your comments!

[Brian Odegaard @BrianOdegaard2]

Really excited for my laboratory's first-ever trip to @Psychonomic_Soc in NYC! (I have also never been, so we are all first-timers). We are presenting three posters with different lines of research from our lab. We would love to chat with many of you about our work! 1/

[Kou Murayama @KouMurayama]

I have been very inactive on social media recently but will try to recover... Here is our new publication on #causalinference and cross-lagged panel models in Psych Metods. People use RI-CLPM as a default nowadays but we should think more deeply about it.

https://t.co/Zl8UFmfC89

[Rob Mok @Rob_Mok]

Thrilled to announce that I will be joining CiNet (Center for Information and Neural Networks; https://t.co/A8jvR5d52y) in Osaka Japan to start my lab in April/May 2025! We have postdoc positions (1 open NOW) & PhD apps (Osaka U) possible. Details below & pls retweet! Short 🧵 1/

[Matthias Michel @MatthiasMichel_]

Excited about this new paper! Suppose you lost half a field of view, or color vision, would you notice? In a new paper in @TrendsCognSci, we show that you probably wouldn’t! You can find out more about those striking failures of visual metacognition here: https://t.co/B3jESVMWTF.

[Kording Lab 🦖 @KordingLab]

When I talk about many rigor-related issues, my stance is not to make *people* look bad. It is not about scientists making mistakes - we always did that. It is about a need to slow down. Fewer papers. More solid. More thinking. It is about the need for science to renew itself.