New work by my former PhD student, Boyang Li His team produced 500 stories of less than 100 words. LLMs were basically chance-level at answering binary questions about the stories arxiv.org/abs/2601.12410

now accepted at ICLR! 🐺🥳🐺 arxiv.org/abs/2506.20666

Very cool idea! Some quick thoughts. It looks like the corruption preserves a lot of information (function words, morphology, word order, punctuation, numbers, register) which would strongly constrains the posterior over plausible discourse frames as it were. 1/3

With @jesusoxford.bsky.social we are looking for a Professor of Statistics. Become part of a historic institution and a community focused on academic excellence, innovative thinking, and significant practical application. About the role: tinyurl.com/b8uy6mr5 Deadline: 15 September

I'm happy to share that I'll be joining Oxford this fall as an associate professor, as well as a fellow of @jesusoxford.bsky.social and affiliate with the Institute for Ethics in AI. I'll also begin my AI2050 Fellowship from @schmidtsciences.bsky.social there. Looking forward to getting started!

Can LLMs reason by analogy like humans? We investigate this question in a new paper published in the Journal of Memory and Language (link below). This was a long-running but very rewarding project. Here are a few thoughts on our methodology and main findings. 1/9

I wrote an entry on Transformers for the Open Encyclopedia of Cognitive Science (‪@oecs-bot.bsky.social‬). I had to work with a tight word limit, but I hope it's useful as a short introduction for students and researchers who don't work on machine learning: oecs.mit.edu/pub/ppxhxe2b

Happy to share this updated Stanford Encyclopedia of Philosophy entry on 'Associationist Theories of Thought' with @ericman.bsky.social. Among other things, we included a new major section on reinforcement learning. Many thanks to Eric for bringing me on board! plato.stanford.edu/entries/asso...

The sycophantic tone of ChatGPT always sounded familiar, and then I recognized where I'd heard it before: author response letters to reviewer comments. "You're exactly right, that's a great point!" "Thank you so much for this insight!" Also how it always agrees even when it contradicts itself.

Despite extensive safety training, LLMs remain vulnerable to “jailbreaking” through adversarial prompts. Why does this vulnerability persist? In a new paper published in Philosophical Studies, I argue this is because current alignment methods are fundamentally shallow. 🧵 1/13

Transformer-based neural networks achieve impressive performance on coding, math & reasoning tasks that require keeping track of variables and their values. But how can they do that without explicit memory? 📄 Our new ICML paper investigates this in a synthetic setting! 🎥 youtu.be/Ux8iNcXNEhw 🧵 1/13

Ah... the morning Australian ritual, waking up and checking into Bluesky with the thought "what fresh hell happened overnight while I was asleep?"

I'm mildly amused by the fact that when you watch obscure videos on syntactic theory, Youtube will serve you ads for Gammarly

I'm back in the Bay Area for this great workshop at UC Berkeley – if you're in the area and interested in LLMs & Cog Sci, come along! simons.berkeley.edu/workshops/ll...

Losing Lynch is a strange feeling. This is hardly an original thing to say, but his work left a big impression on me since I was a teenager. I've rewatched most of his movies over the past year, they're every bit as enthralling as I remembered them. Now it's time to rewatch Twin Peaks!

My article 'Constitutive Self-Consciousness' is now published online in the Australasian Journal of Philosophy. It argues (spoiler alert!) against the claim that self-consciousness is constitutive of consciousness.

I'm happy to share that I'll be one of Schmidt Sciences's new AI2050 fellows! I'll be focusing on addressing the risk of interpretability illusions in AI – cases where interpretability methods yield seemingly plausible yet incorrect explanations. www.schmidtsciences.org/schmidt-scie...

Three ManyBabies projects - big collaborative replications of infancy phenomena - wrapped up this year. The first paper came out this fall. I thought I'd take this chance to comment on what I make of the non-replication result. 🧵 bsky.app/profile/laur...

He's making a list, He's checking it twice He's gonna find out Who's naughty or... Searching in an unsorted list takes linear time, Christmas is postponed to January

Inspired by @mariaa.bsky.social's custom feed for NLP papers, I created a custom feed for philosophy papers posted on BlueSky. It's not perfect, but it works decently well: bsky.app/profile/did:...

Happy to see Bluesky taking off. Here's an attempt at a Philosophy of AI starter pack: go.bsky.app/8pf4odt

I also persist in saying "computational resources" which almost feels old-fashioned. Though I believe that's what linguists call deverbal zero-nominalization and it's pretty common in English, so at some point it'll probably become part of the lexicon?

reinforcement learning

PSA: If you're on Bluesky and Mastodon (or another fediverse network), you can bridge the two with this tool: fed.brid.gy

ICYMI: The School of Computer Science at the University of Sydney 🇦🇺 is hiring, with multiple (equiv.) tenure-track positions. All areas of CS welcome, with a specific focus on Systems, PL, ML, and Quantum Computing. usyd.wd3.myworkdayjobs.com/en-US/USYD_E... Please share, and get in touch!

Graduating PhDs and postdocs in AI, ML, Cogsci, or related areas: apply to work with me and others on AI models of visual and multimodal reasoning. Two years of funding with possible extension to a third year. Application deadline November 22. santafe.edu/about/jobs/p...

New chapter with @cameronbuckner.bsky.social on interventionist methods for interpreting deep neural networks. We review and discuss the significance of recent developments in interpretability research on DNNs from a philosophical perspective. 1/6 philpapers.org/rec/MILIMF-2

Nick Shea's terrific book on concepts is now out, and it is open access. Download it and read it! philpapers.org/rec/SHECAT-11

📄I finally preprinted this new chapter for the upcoming Oxford Handbook of the Philosophy of Linguistics, edited by the excellent Gabe Dupre, Ryan Nefdt & Kate Stanton. Before you get mad about the title – read on! 1/ arxiv.org/abs/2408.07144

New paper (in open access): "Philosophy of cognitive science in the age of deep learning" – in which I argue that although progress in DL is largely driven by engineering goals, it is far from irrelevant to (the philosophy of) cog sci, and vice versa. wires.onlinelibrary.wiley.com/doi/10.1002/...

I look forward to speaking at the 12th Annual Marshall M. Weinberg Symposium @UMich on Friday, alongside researchers whose work I deeply admire – Yejin Choi, @melaniemitchell.bsky.social, and Paul Smolensky. More information including link to the livestream: lsa.umich.edu/weinberginst...

In this new piece for the Institute of Art and Ideas, I examine the messy state of model evaluation in AI research and argue that we need to adopt best practices from cognitive science & supplement behavioral tests with interventions. Read here: iai.tv/articles/the...

Stop the presses, this thing musicologists and composers have been saying for literal decades just got validated by one of the fields of study people actually respect

OpenAI unveiled its video generation model Sora two weeks ago. The technical report emphatically suggests that video generation models like Sora are world simulators. Are they? What does that even mean? I'm taking a deep dive into these questions in a new blog post (link below).

📣📣Delighted to share our paper with amazing team Safoora Yousefi, Leo Betthauser, Hosein Hasanbeig, @raphaelmilliere.com on neuro-inspired analysis of LLMs, LLama2 70B & Vicuna 13B. Decoding In-Context Learning: Neuroscience-inspired Analysis of Representations in LLMs arxiv.org/abs/2310.00313 🧵1/n

There's a lot of speculation about whether OpenAI's video generation model Sora has a 'physics engine' (bolstered by OAI's own claims about 'world simulation'). Like the debate about world models in LLMs, this question is both genuinely interesting and somewhat ill-defined. 🧵1/

📄Now preprinted – Part I of a two-part philosophical introduction to language models co-authored with @cameronbuckner.bsky.social! This first paper offers a primer on language models and an opinionated survey of their relevance to classic philosophical issues. 1/5 arxiv.org/abs/2401.03910

Dear Philosophy of ML folks, as you transition from holidays to Syllabus Scramble Season: Raphael Milliere and I will soon arXiv a paper "A Philosophical Introduction to Language Models", if you are looking for a reading on language models/transformers...

#booklaunch day! Surrounded by family, friends, and colleagues who want to understand AI, but struggle to put the last ten years of furious progress into a systematic, philosophically-grounded framework? I have the perfect stocking-stuffer for you! global.oup.com/academic/pro...

In this new Vox piece written with Charles Rathkopf, we ask why people – experts included – are so polarized about the kinds of psychological capacities they ascribe to language models like ChatGPT, and how we can move beyond simple dichotomies in this debate.

📄 New preprint - The Alignment Problem in Context With the US Executive Order and the UK AI Safety Summit, AI safety concerns have taken the center stage in the past few weeks – But how concerned should we be in light of current developments? 🧵 1/