Wouldn’t it be great to have questions about LM internals answered in plain English? That’s the promise of verbalization interpretability. Unfortunately, our new paper shows that evaluating these methods is nuanced—and verbalizers might not tell us what we hope they do. 🧵👇1/8

This Friday NEMI 2025 is at Northeastern in Boston, 8 talks, 24 roundtables, 90 posters; 200+ attendees. Thanks to goodfire.ai/ for sponsoring! nemiconf.github.io/summer25/ If you can't make it in person, the livestream will be here: www.youtube.com/live/4BJBis...

We've added a quick new section to this paper, which was just accepted to @COLM_conf! By summing weights of concept induction heads, we created a "concept lens" that lets you read out semantic information in a model's hidden states. 🔎

Im excited for NEMI again this year! I’ve enjoyed local research meetups and getting to know others near me working on interesting problems.

🚨 Registration is live! 🚨 The New England Mechanistic Interpretability (NEMI) Workshop is happening Aug 22nd 2025 at Northeastern University! A chance for the mech interp community to nerd out on how models really work 🧠🤖 🌐 Info: nemiconf.github.io/summer25/ 📝 Register: forms.gle/v4kJCweE3UUH...

How do language models track mental states of each character in a story, often referred to as Theory of Mind? We reverse-engineered how LLaMA-3-70B-Instruct handles a belief-tracking task and found something surprising: it uses mechanisms strikingly similar to pointer variables in C programming!

Can we uncover the list of topics a language model is censored on? Refused topics vary strongly among models. Claude-3.5 vs DeepSeek-R1 refusal patterns:

I'll present a poster for this work at NENLP tomorrow! Come find me at poster #80...

Sheridan asks whether the Dual Route Model of Reading that psychologists have observed in humans also appears in LLMs. In her brilliantly simple study of induction heads, she finds that it does! Induction has a Dual Route that separates concepts from literal token processing. Worth reading ↘️

[📄] Are LLMs mindless token-shifters, or do they build meaningful representations of language? We study how LLMs copy text in-context, and physically separate out two types of induction heads: token heads, which copy literal tokens, and concept heads, which copy word meanings.

What will be the linchpin for AI dominance? Read our NSF/OSTP recommendations written with Goodfire's Tom McGrath tommcgrath.github.io, Transluce's Sarah Schwettmann cogconfluence.com, MIT's Dylan Hadfield-Menell @dhadfieldmenell.bsky.social TLDR; Dominance comes from **interpretability** 🧵 ↘️

I'm searching for some comp/ling experts to provide a precise definition of “slop” as it refers to text (see: corp.oup.com/word-of-the-...) I put together a google form that should take no longer than 10 minutes to complete: forms.gle/oWxsCScW3dJU... If you can help, I'd appreciate your input! 🙏

Induction heads are commonly associated with in-context learning, but are they the primary driver of ICL at scale? We find that recently discovered "function vector" heads, which encode the ICL task, are the actual primary mechanisms behind few-shot ICL! arxiv.org/abs/2502.14010 🧵👇

LLMs are known to perpetuate social biases in clinical tasks. Can we locate and intervene upon LLM activations that encode patient demographics like gender and race? 🧵 Work w/ @arnabsensharma.bsky.social, @silvioamir.bsky.social, @davidbau.bsky.social, @byron.bsky.social arxiv.org/abs/2502.13319

Please help amplify ARBOR, a fantastic new research opportunity! If you’d like to start contributing, NDIF is now hosting DeepSeek R1 8B and 70B, open for all researchers to experiment on via our API. Sign up for API access here: login.ndif.us

I'm excited about this new open research initiative! It kind of feels like this is how science is supposed to be done - collaborating and sharing ideas in the open. If you've thought about studying the mechanisms behind R1 & other reasoning models check it out!

DeepSeek R1 shows how important it is to be studying the internals of reasoning models. Try our code: Here @canrager.bsky.social shows a method for auditing AI bias by probing the internal monologue. dsthoughts.baulab.info I'd be interested in your thoughts.

What was the most important machine learning paper in 2024? My Famous Deep Learning Papers list (that I use in teaching) does not include any new ideas from the last year. papers.baulab.info Which single new paper would you add?

More big news! Applications are open for the NDIF Summer Engineering Fellowship—an opportunity to work on cutting-edge AI research infrastructure this summer in Boston! 🚀

The Phase 2 NDIF Pilot is open for a short window. Apply now to get research capacity on Llama 405b. Deadline is December 31. It is not easy to crack open 405b for research, but NDIF solves the key engineering problems for you. Phase 1 powered several very interesting ICLR submissions...

Do you have a great experiment that you want to run on Llama 405b but not enough GPUs? 🚨 #NDIF is opening up more spots in our 405b pilot program! Apply now for a chance to conduct your own groundbreaking experiments on the 405b model. Details: 🧵⬇️

PhD Applicants: remember that the Northeastern Computer Science PhD application deadline is Dec 15. It's a terrific time to do a PhD, with so many interesting things happening in AI. Apply here: www.khoury.northeastern.edu/apply/phd-ap...

New Preprint 🚀 Can diffusion models draw artistic inspiration from nature? 🤔 @huiren and @materzynska trained a diffusion model solely on natural images, excluding artwork from pre-training data ❌🎨 Surprisingly, it can mimic art styles! Curious how it works?👇🧵w/ @davidbau and Antonio Torralba

Do you need to copy art to make art? Hui Ren's and Joanna Materzynska's Art-Free Diffusion tests this question and lets you make "imitation-free" AI art Github: github.com/rhfeiyang/ar... Arxiv: arxiv.org/abs/2412.00176 Website: joaanna.github.io/art-free-dif... X: x.com/materzynska/...