iiinova — where your digital self keeps evolving

Reposted by Shahab Bakhtiari

New paper from the Neurosurgery research team at BCM! 🧵 "A population code for semantics in human hippocampus" led by Melissa Franch. We recorded hippocampus single neurons while people listened to podcasts. www.nature.com/articles/s41...

nature.com

A population code for semantics in human hippocampus - Nature Neuroscience

Franch et al. show that human hippocampal neurons encode the meanings of the words we hear through distributed, context-sensitive population activity that mirrors some features of large language model...

17530

Reposted by Shahab Bakhtiari

Andrew Lampinen @lampinen.bsky.social · 26/09/2026

The phrase "stochastic parrot" is full of sound and fury, signifying nothing. Which, ironically, is exactly what the authors got wrong about language, and the core technical mistake of the paper. 1/ (cross-quote-post because I think this topic is important)

Melanie Mitchell on Twitter
Everyone! This is a straw-person argument. The Stochastic Parrot paper was about LLMs of 2021, not the AI of today, which are not LLMs but complex software systems with vast post training and many external software components.

5543

Reposted by Shahab Bakhtiari

Félicie Dhellemmes @felicie-dhellemmes.bsky.social · 25/09/2026

Proud to present our latest work "Tracking human foragers and their prey reveals adaptive predator-prey dynamics", now hosted on BioRxiv! 🧵 doi.org/10.64898/202... @ralfkurvers.bsky.social @scioi.bsky.social @arc-mpib.bsky.social @alexschakowski.bsky.social @dominikdeffner.bsky.social

26527

Shahab Bakhtiari @shahabbakht.bsky.social · 24/09/2026

I’ve listened to a lot of Ezra Klein, and the Jensen Huang interview was easily one of the worst. Not because Ezra did anything differently, but because Jensen came in with a rehearsed monologue.

380

Reposted by Shahab Bakhtiari

Amirreza Bahramani @amir137.bsky.social · 31/08/2026

Four months ago, US sanctions shut Iranian students out of Neuromatch. So we built our own! 180+ students, 9 TAs, 14 mentors, and an incredible team made NeuroCONNECT happen. Proud to have been part of it. A reminder that sometimes, together, we can break the barriers and injustices imposed on us.

1246

Shahab Bakhtiari @shahabbakht.bsky.social · 24/09/2026

No surprise to those who do interdisciplinary research. Two main reasons: 1) the review panels are most of the time happily disciplinary. 2) the interdisciplinary calls are rare exceptions and quite competitive. So, yes, interdisciplinary is necessary but the funding system isn’t built for it.

170

Reposted by Shahab Bakhtiari

Blake Richards @tyrellturing.bsky.social · 23/09/2026

I’m beyond honoured to be named a 2026 @schmidtsciences.bsky.social Polymath. With US$2.5 million over five years, I’ll be able to pursue an interdisciplinary project to combine quantitative and qualitative data to better capture human neurodiversity. 🧠📈 www.schmidtsciences.org/2026-polymat...

schmidtsciences.org

101248

Shahab Bakhtiari @shahabbakht.bsky.social · 13/09/2026

[grant season rant] Two-year grants combined with perpetual uncertainty are one of the major ingredients of PI misery.

1102

Shahab Bakhtiari @shahabbakht.bsky.social · 09/09/2026

This hits the core of it: real scientific progress isn’t just handing AI a neat puzzle to solve. True science is the slow, decades-long controversies over what "progress" actually looks like in the first place.

140

Shahab Bakhtiari @shahabbakht.bsky.social · 08/09/2026

I’ve been thinking… now that AI is taking on groundbreaking, award-winning problems across science, what are those equivalent problems in #neuroscience? What would one of these models actually have to do to leave us all in awe? 🧠🤖

3101

Shahab Bakhtiari @shahabbakht.bsky.social · 08/09/2026

Great post. What @phillipisola.bsky.social describes below isn’t a matter of possibility anymore, imo. Given the reports on Astra’s success in robotic control, many will pursue it, and it will surely show some success. But … 1/2

161

Shahab Bakhtiari @shahabbakht.bsky.social · 06/09/2026

Astra’s most striking leap across several fronts (eg robotic control) boils down to spatial intelligence. I used to think genuine spatial understanding was impossible for language models without physical real-world interaction. Well … Astra proved that wrong or it's far more than just a mere LLM

2110

Reposted by Shahab Bakhtiari

Kayson Fakhar @kayson.bsky.social · 02/09/2026

🚨 8 months after, this paper found its home at Science Advances. We came up with a generative brain network model based on game theory and found that the human brain is suboptimal both in terms of wiring cost and communication efficiency. Here's why: www.science.org/doi/10.1126/...

45521

Reposted by Shahab Bakhtiari

Andrew Lampinen @lampinen.bsky.social · 14/08/2026

New position piece w/ @tylerbonnen.bsky.social out now in COBS! We suggest that data augmentation is a useful framework for understanding hippocampal contributions to generalization, and offers a path towards more precise modeling: 1/

Data augmentation as a framework for modeling hippocampal contributions to generalization

14814

Reposted by Shahab Bakhtiari

Tom Wallis @tsawallis.bsky.social · 09/08/2026

New preprint! Have you ever wanted to measure psychometric functions in high-dimensional stimulus spaces, but realised that this is infeasible? We present and validate a technique to adaptively estimate psychometric functions for stimulus spaces up to 50 dimensions. www.biorxiv.org/content/10.6...

biorxiv.org

Adaptive experiments in high-dimensional feature spaces: A particle filtering approach

Behavioral experiments are often infeasible when stimulus spaces have many dimensions or when testing time is limited. One way to address this challenge is adaptive stimulus selection, where informati...

36724

Reposted by Shahab Bakhtiari

Shahab Bakhtiari @shahabbakht.bsky.social · 05/08/2026

Our work on curriculum learning in humans and ANNs is now out in PLOS Computational Biology. Thanks to excellent feedback from the reviewers and editor, the final paper now contains some cool new findings.Take a look! 🧠🤖

1202

Shahab Bakhtiari @shahabbakht.bsky.social · 05/08/2026

1202

Reposted by Shahab Bakhtiari

Matteo Carandini @carandinilab.net · 31/07/2026

The apical dendrite looks great, and does wonders in a slice. What does it do in the living cortex? It depends on the cell. A causal spatial role for apical dendrites in the cortex www.biorxiv.org/content/10.6... by @anyiliu.bsky.social with @kenneth-harris.bsky.social and @fedrossi.bsky.social

A cartoon providing a graphical summary of the paper. It is divided in 3 rows.

The top row shows a pyramidal cell and its apical tuft, and shows the two experiments that we ran: optical dendritic pruning (chopping off the apical tuft), and synaptic imaging done in the same neurons where we do somatic imaging.

The middle row shows the results pertaining to neurons that prefer large stimuli (this is visual cortex): pruning reduced their responses, and the receptive fields of apical synapses are far away from the that of the soma.

The bottom row shows the results pertaining to neurons that prefer small stimuli: pruning seemed to do nothing to the responses, and the receptive field of apical synapses are right around that of the soma.

17427

Reposted by Shahab Bakhtiari

Alana Darcher @alanadarcher.bsky.social · 30/07/2026

🎉 *extremely* excited to share our human single neuron + movie dataset is out in #ScientificData! 📄 Human neuron activity during an 83-minute movie from 2,286 neurons and 29 patients 🔗https://www.nature.com/articles/s41597-026-07955-0 1st public dataset from @humansingleneuron.bsky.social!🧵

513937

Reposted by Shahab Bakhtiari

Lukas Muttenthaler @lukasmut.bsky.social · 22/07/2026

🔮At #ICML2026 we presented Attentive Multi-Layer Fusion for Vision Transformers, an efficient method for decoding information from multiple layers of ViTs for downstream task predictions. It’s almost as performant as full fine-tuning but only an inch more expensive than linear probing!

041

Reposted by Shahab Bakhtiari

Nancy Kanwisher @nancykanwisher.bsky.social · 21/07/2026

Exactly @gershbrain.bsky.social! But more importantly: these mice are overtrained on this one task. Maybe stuff gets more broadly distributed across the brain if it is the one thing you do all day every day. And, they are MICE with tiny brains! So, important study, but let's not over-generalize.

0406

Reposted by Shahab Bakhtiari

hakwan lau @hakwan.bsky.social · 21/07/2026

Looking to the brain to improve energy efficiency of AI when we started 3 years ago, i thought if we didn't pull it off in a couple of months, the topic would be soon out of fashion of coz, long distance collab -> we stalled ... but strangely, it isn't totally irrelevant yet. what does this mean?

0144

Shahab Bakhtiari @shahabbakht.bsky.social · 20/07/2026

Mixed feelings from #NeurIPS ACs on the new task of writing initial meta-reviews. Yeah... "it's yet another task for ACs", "why do what authors are supposed to do"... I'm there with you. But on the plus side, it forces ACs to engage earlier and better with discussions. Right?

040

Reposted by Shahab Bakhtiari

Francesco Poli @francescopoli.bsky.social · 13/07/2026

New Preprint! We gave neural networks infant-like curiosity and let them free to explore. They grew brain-like architecture, followed human synaptic development, and even started to display compositional thought. So we think curiosity doesn't just come from a complex mind... it helps us build one 👇

biorxiv.org

Curiosity shapes brain-like architectures and functions

How does complex cognition emerge from simpler underlying processes? We show that two components are sufficient: infant-like curiosity and brain-like biophysical constraints jointly drive the emergenc...

17628

Reposted by Shahab Bakhtiari

Dan Goodman @neural-reckoning.org · 16/07/2026

New preprint (well, very updated). 🤖🧠🧪 We find that an abstract model of neuromodulation lets spiking neural networks perform much better, particularly in challenging noisy environments, using less energy. Relevant to #neuroscience and #neuromorphic computing. 🧵👇 www.biorxiv.org/content/10.1...

biorxiv.org

Neuromodulation enhances the capability and efficiency of spiking neural networks

Spiking neurons underlie the brain’s extreme energy efficiency, and therefore have great potential in neuromorphic computing, although realising this efficiency in practice has proven challenging. We ...

26927

Reposted by Shahab Bakhtiari

Joao Barbosa @jbarbosa.org · 16/07/2026

Great paper! could someone please point me to a paper or two that has seriously pushed for ideas that would be in contradiction with this decoder? I know @benhayden.bsky.social, @pessoabrain.bsky.social et al have done that on bluesky but an actual research paper? Feels like a total straw man

2143

Shahab Bakhtiari @shahabbakht.bsky.social · 16/07/2026

Important paper: "The response profiles of neurons in different brain regions differ enough that a decoder can determine the region to which a particular neuron belongs." Sounds like the perfect antidote to the "everything is everywhere" view.

2294

Reposted by Shahab Bakhtiari

Liu, Shuze @liushuze.bsky.social · 06/07/2026

Do people solve hard problems by first building a mental model, then searching for the best solution? With @gershbrain.bsky.social, our new manuscript suggests a different route: people often start by retrieving cached solutions, and scaffold representations around them. osf.io/preprints/ps...

14319

Shahab Bakhtiari @shahabbakht.bsky.social · 09/07/2026

Many people are complaining about Anthropic’s J-space work, saying it's anthropomorphizing linear algebra, why even bring up consciousness, etc. But honestly, I’m so happy to see a framework that actually puts abstract theories like GWT to the test in-silico and exposes their conceptual limits.

040

Reposted by Shahab Bakhtiari

Samuel Lippl @sflippl.bsky.social · 08/07/2026

Pretraining + fine-tuning powers modern ML, but we lack a theoretical understanding of how pretraining actually shapes downstream learning. In our new @icmlconf.bsky.social paper, we address this gap! 📅 July 9th, Poster #4502 Session 8! 🧵 arxiv.org/pdf/2602.20062

Diagram with "l-order" on the x-axis and "Pretraining dependence" on the y axis. The diagram highlights that only the upper right triangle is a possible region and highlights four distinct regimes: (I) a rich, pretraining independent regime, (II) a lazy, pretraining-dependent regime, (III) lazy, pretraining-independent regime, and (IV) a rich, pretraining dependent regime. Different initialization parameters move us between these regimes.

1193

Reposted by Shahab Bakhtiari

Marlene Cohen @marlenecohen.bsky.social · 07/07/2026

Now out at PNAS - congratulations to Ramanujan Srinath and the team! Ram's ability and willingness to connect all these data, models, and ideas has changed my thinking/interpretation of many findings. I suspect I will keep learning from this study for a long time. www.pnas.org/doi/10.1073/...

pnas.org

The structure of correlated variability reflects task-relevant information in sensory neurons | PNAS

Shared trial-to-trial variability across sensory neurons is reliably reduced when perceptual performance improves, yet this variability is low dime...

13815

Reposted by Shahab Bakhtiari

Andrea Costantino @costantinoai.bsky.social · 12/11/2025

Super excited to share a new preprint! We asked a simple-but-big question: What changes in the brain when someone becomes an expert? Using chess ♟️ + fMRI 🧠 + representational geometry & dimensionality 📈, we ask: 1️⃣ WHAT information is encoded? 2️⃣ HOW is it structured? 3️⃣ WHERE is it expressed? 1/n

26420

Reposted by Shahab Bakhtiari

Eghbal Hosseini @eghbal-hosseini.bsky.social · 07/07/2026

Today at ICML2026, Jack King presents our work with @evfedorenko.bsky.social, linking internal representation trajectories to LLM behavior. 📄 Representational Curvature Modulates Behavioral Uncertainty in LLMs 🔗https://arxiv.org/abs/2604.23985 📍Hall A #1903 | ⏰ 10:30a-12:15p KST Come say hi!

0164

Reposted by Shahab Bakhtiari

Roxana Zeraati @roxana-zeraati.bsky.social · 07/07/2026

Our paper "Neural timescales from a computational perspective" is finally out in @natneuro.nature.com: www.nature.com/articles/s41... We discuss how computational models and methods can distill empirical observations on neural timescales into quantitative, testable theories of brain computation.

nature.com

Neural timescales from a computational perspective - Nature Neuroscience

This review integrates computational approaches to provide a unified view on how data analysis methods, biophysical mechanistic models and machine learning approaches can help to uncover the origins a...

422986

Reposted by Shahab Bakhtiari

Blake Richards @tyrellturing.bsky.social · 07/07/2026

For a long time I felt like there was a hesitation in neuroscience to ascribe anything like a planning function to the various "replay" events we see in the hippocampus, but I think this is rapidly changing. Here's another new entry in human epilepsy patients: www.nature.com/articles/s41... 🧠📈 🧪

nature.com

Human hippocampal ripples coordinate planning sequences and compositional representations in neocortex - Nature Neuroscience

Human hippocampal ripples and replay interact with the prefrontal cortex to update mental representations online, letting the brain compositionally combine familiar elements in new ways for flexible p...

06112

Reposted by Shahab Bakhtiari

Caroline Robertson @carolinerobertson.bsky.social · 12/06/2026

Very excited to share this paper, led by @ajhaskins.bsky.social ! We find that how people look around the world is stable and idiosyncratic — and that these gaze patterns are shaped in part by the conceptual priorities each person brings to a scene. 1/3 www.pnas.org/doi/10.1073/...

38528

Reposted by Shahab Bakhtiari

Dongyan Lin @dongyanl1n.bsky.social · 26/06/2026

(1/n) Thrilled to share my first paper at Meta FAIR! "EgoBabyVLM: Benchmarking Cross-Modal Learning from Naturalistic Egocentric Video Data" 👶 Human infants learn language from sparse, noisy multimodal input. Today's VLMs can't. We built a benchmark + challenge to close that gap. 🧵

13916

Reposted by Shahab Bakhtiari

Bella Fascendini @bellafascendini.bsky.social · 26/06/2026

This work was inspired by @tomerullman.bsky.social's “illusion illusion” paradigm (arxiv.org/abs/2412.18613) where VLMs mistake illusion-like images for genuine optical illusions. In both cases, models respond to what a problem looks like rather than what it actually requires.🔮

arxiv.org

The Illusion-Illusion: Vision Language Models See Illusions Where There are None

Illusions are entertaining, but they are also a useful diagnostic tool in cognitive science, philosophy, and neuroscience. A typical illusion shows a gap between how something "really is" and how some...

35712

Reposted by Shahab Bakhtiari

Bella Fascendini @bellafascendini.bsky.social · 26/06/2026

New paper! w/ @cocoscilab.bsky.social🧵Can large language models reason flexibly, or have they learned what reasoning looks like? We introduce a new paradigm to test this question—the riddle riddle—and find that humans and LLMs show opposite patterns of performance. 📜

619776

Reposted by Shahab Bakhtiari

Steve Fleming @smfleming.bsky.social · 25/06/2026

Lovely to see the full range of excellent commentaries on our BBS article with @matthiasmichel.bsky.social, together with our response, now out: Target article: www.cambridge.org/core/journal... Commentaries: www.cambridge.org/core/journal... Our response: www.cambridge.org/core/journal...

cambridge.org

Sensory horizons and the functions of conscious vision | Behavioral and Brain Sciences | Cambridge Core

Sensory horizons and the functions of conscious vision - Volume 49

25932

Reposted by Shahab Bakhtiari

Amir-massoud Farahmand @sologen.bsky.social · 25/06/2026

Temporal Difference Learning for Diffusion Models (ICML 2026) arxiv.org/abs/2606.15048 By Yangchen Pan (my former PhD student) and co-authors. It reformulates diffusion training as a Markov reward process and introduces a TD objective to encourage temporal consistency across denoising steps.

arxiv.org

Temporal Difference Learning for Diffusion Models

Diffusion models are typically trained with objectives that focus on local denoising targets at individual time steps (or adjacent pairs), which do not enforce consistency between predictions along th...

1152

Reposted by Shahab Bakhtiari

Melanie Walsh @mellymeldubs.bsky.social · 24/06/2026

Excited to share this. @neel2112.bsky.social, @mariaa.bsky.social, and I analyzed 500K anonymous ChatGPT convos (shared w/ consent from WildChat) to see if people were generating fiction. We found tons of stories, fanfiction & erotica. Many users iterated on the same stories for days and weeks.

Screenshot of paper abstract that reads:

AI FICTION IN THE WILD Neel Gupta Maria Antoniak Melanie Walsh

Some professional authors are beginning to use AI tools to help produce their fiction writing. Are readers using AI to generate fiction, too? Drawing on over 500,000 anonymized, English-language ChatGPT-user conversations (Zhao et al.), we find that more than one third of the conversations involve some form of fiction generation—including original stories, roleplay, fanfiction, and erotica. This AI-generated fiction is notably dominated by power users. We identify common fiction generation patterns and profiles among these users, including what we call infinite story demanders, who repeatedly request and revise variations of the same or similar narratives over extended periods of time. We show that users especially gravitate toward fanfiction and erotica, and that they are broadly drawn to generic forms, repetition, immediacy, and niche combinations of story elements. Our findings motivate two theoretical provocations. First, we argue that AI technologies may lead to a shift in the conventional relationship between the author and reader, potentially producing what we call a solipsistic reader-writer, who both generates and consumes fiction within a closed conversational loop, interacting with a machine rather than a human other. Second, we note that LLMs enable interactivity, play, and permutation in ways that are seemingly pleasurable for users, raising questions about where AI will fit into contemporary storytelling and entertainment ecosystems. We situate these developments within broader transformations in literature and media, including self-publishing, fanfiction, and pornography, and suggest that AI-generated fiction shares structural affinities with on-demand, personalized, and repetitive cultural forms.

515645

Reposted by Shahab Bakhtiari

Caswell Barry @caswell.bsky.social · 23/06/2026

Really excited about this work with the awesome Will de Cothi & @shipleysj.bsky.social, out in Nature Reviews Neuroscience. We propose acetylcholine is an analogue of dopamine — tracking not reward prediction errors but state-transition prediction errors. doi.org/10.1038/s41583-026-01058-w

25425