Yet another (popular) article on human/LM reasoning that (as usual) I'm deeply disappointed to see casually presupposes that humans do something called "reasoning" that's implied to be linking together steps that logically follow without any qualification, then proceeds to be dismissive 1/

What should we make of recent language model advancements in mathematics? In my new post, I reflect on long-standing cognitive debates about symbols and neural networks in light of this progress. infinitefaculty.substack.com/p/symbols-ne...

Symbols, neural networks, and mathematical intelligence

Some reflections on connectionism and the basis of higher-level cognition

infinitefaculty.substack.com

Thoughtful and relevant outside math. A partial summary: Gowers thinks it’s important to sustain a human mathematical culture, but is unconvinced by the Leiden Declaration’s attempt to do that by reaffirming human ownership of specific *discoveries*.

tachikoma@tachikoma.elsewhereunbound.com · last wk.

an interesting blog post by Timothy Gowers on the Leiden declaration (on AI and Math), on why he didn't sign. it gets to a subtler aspect of control over AI and our future.

I'm recruiting two postdocs (or potentially PhD students) to work on projects at the intersection of cognition, neuroscience, and AI (with a particular emphasis on mechanistic interpretability). Apply here: docs.google.com/forms/d/e/1F... and please share!

A few months ago (on a break between jobs) I went on the @stanfordpsypod.bsky.social to talk about my research journey from my cognitive psych PhD to industry, differences between academia and industry, what I've worked on, and how industry has changed since I joined. The episode is out now:

CCN began at Columbia in 2017. As #CCN2026 returns to New York City, a special Back2NY event will feature a crowd-sourced panel discussion on how the community has grown, prior challenges, and future perspectives. Share your experiences in the community survey: 2026.ccneuro.org/back2ny/

We make flexible choices in new situations by knitting together information from separate relevant memories. But what governs which memories are retrieved and when? In a new preprint, we captured how people build decision variables from different memories by tracking their gaze on a blank screen.

Flexible decisions arise from resource-rational memory sampling

Flexible decision making depends on retrieving and recombining memories. Yet because this process unfolds covertly, its governing principles remain unknown. Here we use gaze reinstatement to uncover t...

biorxiv.org

Pleased to share that this work is now published in TMLR! openreview.net/forum?id=RuW...

Andrew Lampinen@lampinen.bsky.social · 11mo ago

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:

Now published in open access! Your one-stop shop for the philosophy of language models. It's the spiritual descendant of our two-part preprint from 2024, fully updated. This should be particularly useful for anyone looking for an entry point into this rapidly growing field.

The Philosophy of Language Models

The success of large language models (LLMs) across many domains of AI research has generated intense debate. Some attribute their impressive performance on complex tasks to human-like linguistic and ...

compass.onlinelibrary.wiley.com

What are the real problems to be solved in continual learning? In my latest post, I tackle this question — reviewing where I think the field went astray in the past, how language models changed things, and where the real challenges remain. infinitefaculty.substack.com/p/what-are-t...

What are the real problems of continual learning?

Reflections on catastrophic interference, plasticity, and learning for the future in the era of large language models

infinitefaculty.substack.com

Do LLMs *understand* language? Do educational AI agents *understand* the material they teach (or their students)? Claims about what AI systems do or don't understand are pervasive, but assessing them requires an account of MACHINE UNDERSTANDING

We've updated the preprint of our Naturalistic Computational Cognitive Science paper (arxiv.org/abs/2502.20349) — we've tried to clarify and streamline the arguments, and added some new examples: 1/5

Naturalistic Computational Cognitive Science: Towards generalizable models and theories that capture the full range of natural behavior

How can cognitive science build generalizable theories that span the full scope of natural situations and behaviors? We argue that progress in Artificial Intelligence (AI) offers timely opportunities ...

arxiv.org

🤖🧠NEW PAPER🧠🤖 Children & neural networks can learn syntax from linear strings of words. How do they do it? Our hypothesis: Word co-occurrence statistics provide cues to syntax! (I.e., a new type of bootstrapping to consider!) Paper: arxiv.org/abs/2605.20529 1/n

What is a psychological theory? Here's our take on this tricky and controversial question in this week's Experimentology chapter summary. Many things called "theories" in psychology aren't actually theories — they're frameworks. 🧵 experimentology.io

To appear in Computational Brain & Behavior soon: the claimed 2024 proof (also in CBB) that AGI via learning is intractable also "proves" that ImageNet is intractable. My reading of the hole: equivocation on the variable D. Preprint here: arxiv.org/abs/2411.06498

Barriers to Complexity-Theoretic Proofs that "AGI" Using Machine Learning is Impossible

A recent paper (van Rooij et al. 2024) claims to have proved that achieving human-like intelligence using learning from data is intractable in a complexity-theoretic sense. We point out that the proof...

arxiv.org

Can language models use subtext in their communication? Can they use common ground to incorporate subtext more effectively? In our new preprint, we study these questions across various domains — from visual communication to story writing games.

When and how can test-time thinking allow models to use information latent in their training data? What are the benefits and tradeoffs relative to other solutions like synthetic data augmentation? Pleased to share (after a long delay) an exploration of these issues: arxiv.org/abs/2604.01430 thread:

Improving Latent Generalization Using Test-time Compute

Language Models (LMs) exhibit two distinct mechanisms for knowledge acquisition: in-weights learning (i.e., encoding information within the model weights) and in-context learning (ICL). Although these...

arxiv.org

Pleased to share that our paper "Representation Biases: Variance is Not Always a Good Proxy for Importance" is now out as Theory/New Concepts paper in eNeuro! www.eneuro.org/content/13/3... 1/

Representation Biases: Variance Is Not Always a Good Proxy for Importance

A central approach in neuroscience is to analyze neural representations as a means to understand a system's function, through the use of methods like principal component analysis, regression, and repr...

eneuro.org

I joined Anthropic (alignment team) this week — exciting place to be at an exciting time!

Sharing “Neural Thickets”. We find: In large models, the neighborhood around pretrained weights can become dense with task-improving solutions. In this regime, post-training can be easy; even random guessing works Paper: arxiv.org/abs/2603.12228 Web: thickets.mit.edu 1/

What is the interplay between representations learned from (language) surface forms alone, and those learned from more grounded evidence (e.g.,vision)? Excited to share new work understanding “Cross-modal taxonomic generalization” in (V)LMs arxiv.org/abs/2603.07474 1/

Short post on what I call the "no-magic approach to understanding intelligent systems" — the philosophy I think of as motivating our work on understanding intelligence without resorting to magical thinking about AI or humans! infinitefaculty.substack.com/p/the-no-mag...

The no-magic approach to understanding intelligent systems

Today I want to write a bit about the philosophy I think underlies much of the work that my collaborators and I (as well as many other researchers that I respect) have done on understanding artificial...

infinitefaculty.substack.com

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!

tyler bonnen@tylerbonnen.bsky.social · 5mo ago

excited to share some recent work! neural networks trained on multi-view sensory data are the first to match human-level 3D shape perception we predict human accuracy, error patterns, and reaction time—all zero-shot, no training on experimental data arxiv.org/abs/2602.17650 1/🧠

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...

Dileep George joins Astera to lead its neuro-inspired AGI effort

Dileep George is joining Astera as Head of AI, leading our AGI research division. Working alongside our Chief Scientist Doris Tsao, he and the team will explore novel, brain-inspired computational arc...

astera.org

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.

Memorization vs. generalization in deep learning: implicit biases, benign overfitting, and more

Or: how I learned to stop worrying and love the memorization

infinitefaculty.substack.com

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!

Open Encyclopedia of Cognitive Science bot@oecs-bot.bsky.social · 6mo ago

AI Model Evaluation by Jennifer Hu: https://doi.org/10.21428/e2759450.b0757e8d