Aaron Mueller

14 Aug

NEMI workshop is starting now! Couldn’t make it to Boston? Check out the livestream on our website

Aaron Mueller reposted

5 Aug

🚨 NEMI decisions are out! Be sure to check your spam folder for the decision email, as a few have ended up there. Looking forward to seeing you at NEMI! 🎉

27 July

Tired of writing NeurIPS rebuttals? Take a break by registering to attend the New England Mechanistic Interpretability workshop (deadline today)!

New England Mechanistic Interpretability Workshop@nemiworkshop.bsky.social

The NEMI workshop registration deadline is today, July 27th! Last chance to sign up to meet New England’s mech interp community, including Rhett the Terrier, who’s using J-Space to learn what LLMs think when he begs them for treats. Link in thread 👇

Aaron Mueller reposted

15 July

The NEMI crew are entering the boat parade at Sail Boston while we wait for the mech interp abstracts to sail in on August 1

Aaron Mueller reposted

14 July

NEMI is exactly one month away! Get your abstracts in by August 1st

Micah Benson@micahben.bsky.social

🧠🤖 The 2026 New England Mechanistic Interpretability (NEMI) Workshop will be Aug. 14 at Boston University! Help spread the word and join the New England mech interp community! Registration and submission info in thread:👇

If you'll be at ACL or ICML this year, come check out the work from our group and collaborators - summary 🧵 below. Lots to like for those into {mechanistic, developmental, pragmatic} interpretability! I'll be at ACL; say hi!

What kinds of interpretability claims are licensed by current methods? We argue that causality provides a useful framework for matching evidence to claims, and clarifying when interpretability conclusions generalize. Chat with @shrutijoshi.bsky.social at ICML!

How do language models update entity states? Many findings, but one highlight is that removing an entity locally often causes a model to remove it everywhere. We apply this understanding to improve performance. Chat with Peter Tang at ICML!

Most interpretability is post hoc. How can we understand when a feature is learned during training? We apply crosscoders to understand when models learn syntactic features, and when multilinguality arises. Chat with @bayazitdeniz.bsky.social at ACL! bsky.app/profile/baya...

Deniz Bayazit@bayazitdeniz.bsky.social

1/🚨 New preprint How do #LLMs’ inner features change as they train? Using #crosscoders + a new causal metric, we map when features appear, strengthen, or fade across checkpoints—opening a new lens on training dynamics beyond loss curves & benchmarks. #interpretability

We show that the benefits of multi-agent debate can be distilled into a single model. Our method IMAD improves performance at a fraction of the tokens, and can improve persona steering. Learn more from John Seon Keun Yi at ACL!

SAEs give us fine-grained control over LLMs. How can we permanently encode feature ablations into an LM's parameters? We propose CRISP, and show that this improves unlearning over the prior state-of-the-art. Chat with @tomerashuach.bsky.social at ACL!

How well do interpretability methods disentangle concepts? We propose a multi-concept evaluation setting. Lots of findings, but most boil down to "establishing the independence of two LLM mechanisms *requires* interventional evidence."

nemiconf.github.ioThe 3rd New England Mechanistic Interpretability (NEMI) Workshop

10 June

The New England Mechanistic Interpretability (NEMI) workshop is coming to BU on Aug. 14! Join us for talks, a panel, food, and plenty of opportunities to connect with the many great researchers in the area. Register and help spread the word!

Micah Benson@micahben.bsky.social

Aaron Mueller reposted

Naomi Saphra

@nsaphra.bsky.social

9 June

✨ it's coming ✨ NEMI 2026 will be lit. It will also be the new BU interp supergroup's debut ball. Come meet us!

Aaron Mueller reposted

Computational Linguistics Journal

@complingjournal.bsky.social

13 May

Interpretability provides a toolset for understanding how and why LMs behave in certain ways. This survey proposes a perspective on interpretability research grounded in causal mediation analysis: doi.org/10.1162/COLI... #NLProc #CLJournal @jannikbrinkmann.bsky.social @amuuueller.bsky.social

Aaron Mueller reposted

Micah Benson

@micahben.bsky.social

25 Mar

I truly believe the rapid advances in the mech interp subfield have something real to offer AI ethics researchers: A chance to look beyond the HOW of evals to the WHY, a first pass at a technical solution when we see the opportunity, a new avenue for showing failures that prove models are not gods

tedlab.mit.edutedlab.mit.edu

21 Jan

Representation steering is now a common way to mitigate LLM shortcuts. How much legitimate knowledge does this tend to remove? Turns out that these methods can be surprisingly precise! But also: no single steering operation will fix all shortcuts. Led by @shanzzyy.bsky.social!

Zhengyang Shan@shanzzyy.bsky.social

Can steering remove LLM shortcuts without breaking legitimate LLM capabilities? In our @eaclmeeting.bsky.social paper, we show that conceptual bias is separable from concept detection; this means inference-time debiasing is possible with minimal capability loss.

Aaron Mueller reposted

languagemit.bsky.social

@languagemit.bsky.social

24 Dec 2025

New book! I have written a book, called Syntax: A cognitive approach, published by MIT Press. This is open access; MIT Press will post a link soon, but until then, the book is available on my website: tedlab.mit.edu/tedlab_websi...

Aaron Mueller reposted

Najoung Kim

@najoung.bsky.social

19 Nov 2025

Replying to @najoung.bsky.social

I also want to mention that the lang x computation research community at BU is growing in an exciting direction, especially with new faculty like @amuuueller.bsky.social, @anthonyyacovone.bsky.social, @nsaphra.bsky.social, & @profsophie.bsky.social! Also, Boston is quite nice :)

In LLMs, concepts aren’t static: they evolve through time and have rich temporal dependencies. We introduce Temporal Feature Analysis (TFA) to separate what's inferred from context vs. novel information. A big effort led by @ekdeepl.bsky.social, @sumedh-hindupur.bsky.social, @canrager.bsky.social!

Can@canrager.bsky.social

Humans and LLMs think fast and slow. Do SAEs recover slow concepts in LLMs? Not really. Our Temporal Feature Analyzer discovers contextual features in LLMs, that detect event boundaries, parse complex grammar, and represent ICL patterns.

I'm glossing over our deeper motivations from neuroscience (predictive coding) and linguistics here, but we believe there's significant cross-field appeal for those interested in intersections of cog sci, neuroscience, and machine learning!

arxiv.orgPriors in Time: Missing Inductive Biases for Language Model InterpretabilityRecovering meaningful concepts from language model activations is a central aim of interpretability. While existing feature extraction methods aim to identify concepts that are independent directions,...

Check out the paper and our demo features! 📜 Preprint: arxiv.org/abs/2511.01836 🧠 Play with temporal feature analysis on Neuronpedia: www.neuronpedia.org/gemma-2-2b/1...

Can@canrager.bsky.social

TFA is designed to capture context-sensitive information. Consider parsing: SAE often assign each word to its most frequent syntactic category, regardless of context. Meanwhile, TFA recovers the correct parse given the context!

INTERPLAY Workshop@COLM '25

Jeff Elman famously showed us in 1990 that time is a rich signal in itself. Our work demonstrates that this lesson applies equally well to interpretability methods. The inductive biases of interp methods should reflect the structure of what is being studied.

Aaron Mueller reposted

@interplay-workshop.bsky.social

9 Oct 2025

✨ The schedule for our INTERPLAY workshop at COLM is live! ✨ 🗓️ October 10th, Room 518C 🔹 Invited talks from @sarah-nlp.bsky.social John Hewitt @amuuueller.bsky.social @kmahowald.bsky.social 🔹 Paper presentations and posters 🔹 Closing roundtable discussion. Join us in Montréal! @colmweb.org

What's the right unit of analysis for understanding LLM internals? We explore in our mech interp survey (a major update from our 2024 ms). We’ve added more recent work and more immediately actionable directions for future work. Now published in Computational Linguistics!

Thanks again to the co-authors! Such a wide survey required a lot of perspectives. @jannikbrinkmann.bsky.social Millicent Li, Samuel Marks, @koyena.bsky.social @nikhil07prakash.bsky.social @canrager.bsky.social (1/2)

Aruna Sankaranarayanan, @arnabsensharma.bsky.social absensharma.bsky.social @ericwtodd.bsky.social ky.social @davidbau.bsky.social u.bsky.social @boknilev.bsky.social (2/2)

We also made the causal graph formalism more precise. Interpretability and causality are intimately linked; the latter makes the former more trustworthy and rigorous. This formal link should be strengthened in future work.

arxiv.orgThe Quest for the Right Mediator: Surveying Mechanistic Interpretability Through the Lens of Causal Mediation AnalysisInterpretability provides a toolset for understanding how and why neural networks behave in certain ways. However, there is little unity in the field: most studies employ ad-hoc evaluations and do not...

See the paper for more details! arxiv.org/abs/2408.01416

One of the bigger changes was establishing criteria for success in interpretability. What units of analysis should you use if you know what you’re looking for? If you *don’t* know what you’re looking for?

Aaron Mueller reposted

Andrew Lampinen

@lampinen.bsky.social

5 Aug 2025

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:

bsky.app@nikhil07prakash.bsky.socialHow 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!

17 July 2025

If you're at #ICML2025, chat with me, @sarah-nlp.bsky.social, Atticus, and others at our poster 11am - 1:30pm at East #1205! We're establishing a 𝗠echanistic 𝗜nterpretability 𝗕enchmark. We're planning to keep this a living benchmark; come by and share your ideas/hot takes!

Aaron Mueller reposted

David Bau

@davidbau.bsky.social

25 June 2025

The new "Lookback" paper from @nikhil07prakash.bsky.social‬ contains a surprising insight... 70b/405b LLMs use double pointers, akin to C programmers' double (**) pointers. They show up when the LLM is "knowing what Sally knows Ann knows", i.e., Theory of Mind. bsky.app/profile/nik...

SAEs have been found to massively underperform supervised methods for steering neural networks. In new work led by @danaarad.bsky.social, we find that this problem largely disappears if you select the right features!

Dana Arad@danaarad.bsky.social

Tried steering with SAEs and found that not all features behave as expected? Check out our new preprint - "SAEs Are Good for Steering - If You Select the Right Features" 🧵

By limiting steering to output features, we recover >90% of the performance of the best supervised representation-based steering methods—and at some locations, we outperform them!

We still have a lot to learn in editing NN representations. To edit or steer, we cannot simply choose semantically relevant representations; we must choose the ones that will have the intended impact. As @peterbhase.bsky.social found, these are often distinct.

Dana Arad@danaarad.bsky.social

Tried steering with SAEs and found that not all features behave as expected? Check out our new preprint - "SAEs Are Good for Steering - If You Select the Right Features" 🧵

We define the notion of an “output feature”, whose role is to increase p(some token(s)). Steering these gives better results than steering “input features”, whose role is to attend to concepts in the input. We propose fast methods to sort features into these categories.

Aaron Mueller reposted

Dana Arad

@danaarad.bsky.social