Jin Fan, PhD

Professor of Psychology

[Jin Fan, PhD @jinfanphd]

AI fails classic attention test, via @EurekAlert! @PNASNexus https://t.co/ntp82ORF37

[Suketu Patel @SuketuPatel23]

Question today at #CNS2025, “What do you think AI is missing in order to model biological intelligence?”

My poster answers this: Transformer based LLMs lack a mechanism for cognitive control.

Specifically an executive control of attention to resolve conflicts and interference.

[Suketu Patel @SuketuPatel23]

🚨New pre-print!
Deficient Executive Control in Transformer Attention

Current transformer attention isn't "all you need".

We had ChatGPT 4o & Sonnet 3.5 take the classic Stroop task, testing their executive control of attention Can LLMs handle conflicting information? 1/

[Jin Fan, PhD @jinfanphd]

https://t.co/DninhCF9rs

[Suketu Patel @SuketuPatel23]

🚨AI Alignment pre-print paper!🚨

What Is the Alignment Problem in Artificial Intelligence?
It ensures AI enacts benevolently while having the rationality to foresee harmful consequences. This involves designing AI with an understanding of human values, morals, and ethics. 1/

[Jin Fan, PhD @jinfanphd]

posted on bioRxiv: https://t.co/AHBwDYU5So

[Hominid Dan 🦊 @hominidan]

@pursuitofprog This does make sense to me given human cognitive control only accounts for ~3.5 bps whereas the whole brain activity is estimated to be around 11 million bps and that's not accounting for cognition happening through cells other than neurons https://t.co/qUiURmzoUW

[Sharon Gilaie-Dotan @ShaGilDot]

(1/) On Tues Prof James Haxby @haxbylab will give a talk on "Modeling shared and variable information encoded in fine-scale cortical topographies"
@ https://t.co/Z5pH14vX5K
@worldwideneuro #BIUVisionScienceSeminar
@dartmouth @ubarilan @GondaBrain
#vision #neuroscience #science

[Eiko Fried @EikoFried]

Preprint finds in n=9000 that emotional dot probe task is “not an adequately reliable measure of individual differences in attentional bias to threat” & doesn’t lend itself to “quantify differences between people, incl. individuals w elevated anxiety.”

https://t.co/TGhS6jady0

[Alfredo Spagna: alfredospagna.bsky.social @SpagnaPhD]

Although the «frontoparietal network (FPN) was commonly activated for both overt and covert orienting, —-> multivariate patterns of the activation of voxels in the FPN accurately predicted whether eye movements were involved or not during orienting. <— 🔥🧠👍🏼

[Jin Fan, PhD @jinfanphd]

@postlelab Basically they are the same with different terms. Here we used a structural based term (FPN) rather than a functional defined term (DAN).

[Jin Fan, PhD @jinfanphd]

Representational Coding of Overt and Covert Orienting of Visuospatial Attention in the Frontoparietal Network
https://t.co/k223hZGzTx

[Jin Fan, PhD @jinfanphd]

Testing a Cognitive Control Model of Human Intelligence | Scientific Reports #CognitiveControl, #IQ https://t.co/3Vb6HI7aFs

[Jin Fan, PhD @jinfanphd]

Thanks Melissa for making the summary!

[Jin Fan, PhD @jinfanphd]

Accessing the development and heritability of the capacity of cognitive control https://t.co/6D83kYWsj5 @SpagnaPhD @MelAnnMackie #CognitiveControl

[Jin Fan, PhD @jinfanphd]

@QC_News @GC_CUNY

[Jin Fan, PhD @jinfanphd]

Use social media to promote your research

[Jin Fan, PhD @jinfanphd]

#CUNYProud