Can large language models *introspect*? In a new paper, @kmahowald.bsky.social and I study the MECHANISM of introspection in big open-source models. tldr: Models detect internal anomalies through DIRECT ACCESS, but don't know what the anomalies are. And they love to guess “apple” 🍎

After 5.5 years (or 7 or 9, counting internships), today was my last day at Google/DeepMind. When I was in London recently, I walked through the two floors that were (most of) DeepMind when I first joined, and thought about how much the company and field have changed since then.

🚨New preprint! In-context learning underlies LLMs’ real-world utility, but what are its limits? Can LLMs learn completely novel representations in-context and flexibly deploy them to solve tasks? In other words, can LLMs construct an in-context world model? Let’s see! 👀

Really cool work — learning over sequential experiences that contain the embodied cue of viewpoint as well as visual inputs, can give rise to human-like 3D shape perception!

News! I've joined the Astera Institute to lead its neuroscience based AGI research. Backed by $1B+ commitment over the coming decade, my team will explore novel, brain-inspired architectures and algos toward safe, efficient human-like AGI, working alongside Doris Tsao. 1/ astera.org/dileep-georg...

What is the relationship between memorization and generalization in AI? Is there a fundamental tradeoff? In infinitefaculty.substack.com/p/memorizati... I’ve reviewed some of the evolving perspectives on memorization & generalization in machine learning, from classic perspectives through LLMs.

Excited to launch Principia, a nonprofit research organisation at the intersection of deep learning theory and AI safety. Our goal is to develop theory for modern machine learning systems that can help us understand complex network behaviors, including those critical for AI safety and alignment. 1

I wrote a short article on AI Model Evaluation for the Open Encyclopedia of Cognitive Science 📕👇 Hope this is helpful for anyone who wants a super broad, beginner-friendly intro to the topic! Thanks @mcxfrank.bsky.social and @asifamajid.bsky.social for this amazing initiative!

This work by @mariaeckstein.bsky.social et al is a nice example of how progress in psychology can be expedited with machine learning. How long before this type of approach is expected for models-of-behavior papers? My guess: not long. (If you are a trainee, nudge!) www.nature.com/articles/s41...

The visual world is composed of objects, and those objects are composed of features. But do VLMs exploit this compositional structure when processing multi-object scenes? In our 🆒🆕 #ICLR2026 paper, we find they do – via emergent symbolic mechanisms for visual binding. 🧵👇

Interesting results by @eghbal-hosseini.bsky.social on how language models representation geometry evolves during different types of in-context learning!

Was a pleasure to discuss the cognitive basis of reasoning at an @ivado.bsky.social workshop with legends like @alisongopnik.bsky.social @lauraruis.bsky.social @taylorwwebb.bsky.social and Andrew Granville!

New paper! In arxiv.org/abs/2601.20834 we study how language models representations of things like factuality evolve over a conversation. We find that in edge case conversations, e.g. about model consciousness or delusional content, model representations can change dramatically! 1/

Should you go to academia or industry for research in AI or cognitive science? It's the most common question I get asked by PhD students, and I've written up some of my thoughts on the answer, as an epilogue to my research-focused series on these fields: infinitefaculty.substack.com/p/on-researc...

Our experiences have countless details, and it can be hard to know which matter. How can we behave effectively in the future when, right now, we don't know what we'll need? Out today in @nathumbehav.nature.com , @marcelomattar.bsky.social and I find that people solve this by using episodic memory.

I'm very excited about this paper with @yngwienielsen.bsky.social just out in @nathumbehav.nature.com in which we provide evidence for the mental representation of non-hierarchical linguistic structure in language use. 🧵 1/4 Read the paper here: rdcu.be/eZ26u

arxiv.org/abs/2601.11432 I want to share an astonishing result. LLMs can "translate" Jabberwocky' texts like 'He dwushed a ghanc zawk” & even and even 'In the BLANK BLANK, BLANK BLANK has BLANK over any BLANK BLANK’s BLANK' This has profound consequence for thinking about.. 1/2

When are impossibility proofs misleading? In infinitefaculty.substack.com/p/be-wary-of..., I discuss a common issue I see: proofs that are logically valid, but where the underlying assumptions are unjustified. I discuss ‘proofs’ that cognition cannot be tractably learned, and that LMs are 1/

What can cognitive science learn from AI? In infinitefaculty.substack.com/p/what-cogni... I outline how AI has found that scale and richness of learning experiences fundamentally change learning & generalization — and how I believe we should rethink cognitive experiments & theories in response.

New post! Last week I shared why I thought cognitive (neuro)science hasn’t contributed as much as one might hope to the design of AI systems; this week I'm sharing my thoughts on how methods and principles from these fields *have* been useful in my work. infinitefaculty.substack.com/p/how-cognit...

Why isn’t modern AI built around principles from cognitive science or neuroscience? Starting a substack (infinitefaculty.substack.com/p/why-isnt-m...) by writing down my thoughts on that question: as part of a first series of posts giving my current thoughts on the relation between these fields. 1/3

I'm more and more convinced that low-dimensional manifolds in the brain are just an artifact of the experimental designs and analyses we use... 🧠📈 🧪

Dimensionality reduction may be the wrong approach to understanding neural representations. Our new paper shows that across human visual cortex, dimensionality is unbounded and scales with dataset size—we show this across nearly four orders of magnitude. journals.plos.org/ploscompbiol...

Heading to NeurIPS this week! Let me know if you want to chat about science of what models learn, cognitive science, interpretability, what models learn in context vs. from their training data, etc. A few things I'm involved in:

Amazing opportunity to work with a brilliant researcher and all-around wonderful person — definitely apply if you're interested in memory & perception at the intersection of AI & cognitive (neuro)science!

Very important point! We've made arguments from a computational perspective that low-variance features can be computationally relevant (bsky.app/profile/lamp...), but it's much cooler to see it demonstrated on a model of real neural dynamics

Great work by Andrea & co, now out with more datasets, models, and analyses!

I was honored to speak at Princeton’s symposium on The Physics of John Hopfield: Learning & Intelligence this week. I sketched out a perspective that ties together some of our recent work on ICL vs. parametric learning, and some possible links to hippocampal replay: 1/

Can't tell you how great it is to finally be able to release and talk about this work, SIMA 2, the next step toward embodied intelligence in rich, interactive 3D worlds! deepmind.google/sima

Today in Nature Machine Intelligence, Kazuki Irie & I discuss 4 classic challenges for neural nets — systematic generalization, catastrophic forgetting, few-shot learning, & reasoning. We argue there is a unifying fix: the right incentives & practice. rdcu.be/eLRmg

What aspects of human knowledge do vision models like CLIP fail to capture, and how can we improve them? We suggest models miss key global organization; aligning them makes them more robust. Check out LukasMuttenthaler's work, finally out (in Nature!?) www.nature.com/articles/s41... + our blog! 1/3

New work to appear @ TACL! Language models (LMs) are remarkably good at generating novel well-formed sentences, leading to claims that they have mastered grammar. Yet they often assign higher probability to ungrammatical strings than to grammatical strings. How can both things be true? 🧵👇

Apologies for being quiet on here lately — been focusing on the more important things in life :)

Pleased to share that our survey "Getting aligned on representational alignment" — on representational alignment across cognitive (neuro)science and machine learning — is now published in TMLR! openreview.net/forum?id=Hiq... Kudos to @sucholutsky.bsky.social @lukasmut.bsky.social for leading this!

🚨 New preprint alert! 🧠🤖 We propose a theory of how learning curriculum affects generalization through neural population dimensionality. Learning curriculum is a determining factor of neural dimensionality - where you start from determines where you end up. 🧠📈 A 🧵: tinyurl.com/yr8tawj3

Why does AI sometimes fail to generalize, and what might help? In a new paper (arxiv.org/abs/2509.16189), we highlight the latent learning gap — which unifies findings from language modeling to agent navigation — and suggest that episodic memory complements parametric learning to bridge it. Thread:

How can an imitative model like an LLM outperform the experts it is trained on? Our new COLM paper outlines three types of transcendence and shows that each one relies on a different aspect of data diversity. arxiv.org/abs/2508.17669

In neuroscience, we often try to understand systems by analyzing their representations — using tools like regression or RSA. But are these analyses biased towards discovering a subset of what a system represents? If you're interested in this question, check out our new commentary! Thread:

Get ready to enter the simulation... Genie 3 is a new frontier for world models: its environments remain largely consistent for several minutes, with visual memory extending as far back as 1min. These limitations will only decrease with time. Welcome to the future.🙌 deepmind.google/discover/blo...

Looking forward to attending CogSci this week! I'll be giving a talk (see below) at the Reasoning Across Minds and Machines workshop on Wednesday at 10:25 AM, and will be around most of the week — feel free to reach out if you'd like to meet up!

This summer my lab's journal club somewhat unintentionally ended up reading papers on a theme of "more naturalistic computational neuroscience". I figured I'd share the list of papers here 🧵:

Does vision training change how language is represented and used in meaningful ways?🤔The answer is a nuanced yes! Comparing VLM-LM minimal pairs, we find that while the taxonomic organization of the lexicon is similar, VLMs are better at _deploying_ this knowledge. [1/9]

Quick thread on the recent IMO results and the relationship between symbol manipulation, reasoning, and intelligence in machines and humans:

Excited to share a new project spanning cognitive science and AI where we develop a novel deep reinforcement learning model---Multitask Preplay---that explains how people generalize to new tasks that were previously accessible but unpursued.

Exciting new preprint from the lab: “Adopting a human developmental visual diet yields robust, shape-based AI vision”. A most wonderful case where brain inspiration massively improved AI solutions. Work with @zejinlu.bsky.social @sushrutthorat.bsky.social and Radek Cichy arxiv.org/abs/2507.03168

Thrilled to see our TinyRNN paper in @nature! We show how tiny RNNs predict choices of individual subjects accurately while staying fully interpretable. This approach can transform how we model cognitive processes in both healthy and disordered decisions. doi.org/10.1038/s415...

Experimentology is out today!!! A group of us wrote a free online textbook for experimental methods, available at experimentology.io - the idea was to integrate open science into all aspects of the experimental workflow from planning to design, analysis, and writing.

Really nice analysis!

Humans and animals can rapidly learn in new environments. What computations support this? We study the mechanisms of in-context reinforcement learning in transformers, and propose how episodic memory can support rapid learning. Work w/ @kanakarajanphd.bsky.social : arxiv.org/abs/2506.19686

I'm excited about this TICS Opinion with @yngwienielsen.bsky.social, challenging the view that structural priming—the tendency to reuse a recent syntactic structure—provides evidence for the psychological reality of grammar-based constituent structure. authors.elsevier.com/a/1lIFK4sIRv... 🧵1/4