Ashish Kharbanda·6d agoMechanistic Interpretability of LLM Refusal BehaviourWhere does a language model keep its “refusals”?A response icon1A response icon1
InTowards AIbyVedant Pandhare·Nov 3Your Model Passed Safety Evals. Then You Quantized the KV Cache.You compressed your way into a vulnerability nobody was checking for.A response icon1A response icon1
Subisu·Nov 1Can We Mute a Prompt Injection? Projecting Out Role Confusion in an LLMCore idea, in one line
Adele Chinda·Oct 29Can we read a model’s mind? What probing actually tells us (and what it does not)When researchers say a model “knows” something, what do they mean? Usually they mean something more specific than the model getting the…
Gabrieljearle·Oct 29Interpretability Through The Lens of Fractal Plants | Gabriel EarleWe’re now several years into the era where an artificial image, video, or piece of text can be made on command, and it looks startlingly…A response icon1A response icon1
Atharva·Sep 5Are Political Probes Polyglots?Testing cross-language transfer across six languages, four models, and eight activation-extraction methodsA response icon1A response icon1
Adnan Masood, PhD.·Aug 22Getting Started with Mechanistic Interpretability — Part 0: Why Mech Interp, and Why NowA practitioner’s guide to mech interp: the evidence for sparse autoencoders, induction heads, and circuit tracing, the negative results…
Narendra Bayutama Wibisono·Aug 13Reading LLMs’ Minds: Disentangling Empathy and Clinical Spaces through Orthogonal…Disentangling Empathy and Clinical Spaces through Orthogonal Auto-Interpretability
Veda Revankar·Aug 12Reconstruction-Downstream Disconnect: Why Interpretable Features Aren’t Always CausalMechanistic interpretability has long sought to turn the “black box” of neural networks into a white-box system we can truly understand…
