MLn Club (ML Reading Group) #15: Interpreting Language Model Parameters
Welcome to Week 15: Interpreting Language Model Parameters
Can a model’s learned mechanisms be recovered directly from its parameters, without assuming that neurons, attention heads, or activation features are the correct units of analysis?
Goodfire’s adVersarial Parameter Decomposition (VPD) attempts to interpret language models by decomposing their actual weight matrices into simple rank-one subcomponents. Unlike activation-based methods such as sparse autoencoders and transcoders, VPD preserves the model’s original computational form: its components sum back to the model’s parameters and are trained so that only a small subset is necessary for any given input. Crucially, VPD adversarially searches for combinations of supposedly unimportant components whose removal would break the model, providing a stronger test of mechanistic faithfulness than ordinary random ablations.
Applied to a four-layer, 67M-parameter language model, VPD discovers interpretable mechanisms that frequently cut across conventional architectural boundaries—including attention computations distributed across multiple heads. These components can be assembled into attribution graphs for behaviors such as pronoun prediction and bracket closing, and the authors demonstrate a preliminary hand-written parameter edit that predictably changes emoticon completions. The result challenges the field’s focus on activation features: rather than asking only what representations a model contains, VPD offers a path toward understanding the parameter-level algorithms that operate on those representations—while leaving open whether the method can scale to frontier models or uniquely recover their “true” mechanisms.
Join us at CASI for discussion at 8 pm, and (optional) quiet reading from 7 pm.
📖 Reading Recommendations, Questions, or Comments? Contact us here!
🔎 View past meeting notes here.
What's this?
A super warm group of folks discussing their favorite topics!
In the first half, we host an optional quiet reading space
In the second half, we have a discussion where people can talk about what they found interesting about the reading and ask questions about things they didn't understand
When/Where:
CMU AI Safety Initiative's Office, 201 Craig Street, right across the PNC bank. Look for the open door up the stairs.
8pm discussion, 7pm optional quiet reading time.
Here's how it usually goes:
7:00 PM — arrival and settling in
8:00 PM — introductions
8:10 PM — discussion time
9:00 PM — wrap up then open discussion
Who's it for?
People who've been wanting to read up on the latest papers in ML and other fields but just haven't been able to find the time/motivation.
Why:
We've been procrastinating too much on our readings, even though we have so much fun doing them. We know we're not alone in this and want to keep others accountable for learning more about what they're passionate about!
We've also met a ton of really fun friends by discussing what we care about!
Rules/guidelines on how to act:
Act like a host, include people in conversations, talk to people even if they're strangers, offer to explain what you know, and keep an open mind! come to read stuff and find super fun friends :)
Bring snacks if you're feeling kind!