The Flow has not summarised this story yet — read it at arXiv Machine Learning.
arXiv:2606. 11722v1 Announce Type: cross Abstract: Finding interpretable directions in language-model representations is critical for understanding and controlling model behavior.
arXiv:2511. 09432v2 Announce Type: replace Abstract: Machine learning (ML) models achieve remarkable performance but remain hard to interpret due to their scale and complexity.
The paper introduces a controlled auditing framework using parametric CAD to separate failures in encoding, expression, and activation-level control in frozen large language models (LLMs). Applied to eight pretrained models and random controls, it finds that pretraining mainly improves linear decoding of local relations, while global constraint status is already highly decodable from random representations. Despite strong local decodability, the models’ zero‑shot predictions lag far behind supervised probe performance, and simple activation steering fails to provide reliable control; however, low‑rank fine‑tuning with only 400 labeled relation pairs can substantially improve generation on three models, indicating that the representation–behavior gap can be narrowed through adaptation.
The paper introduces eXplaining to Learn (eX2L), an interpretable framework that regularizes a classifier by penalizing similarity between Grad‑CAM maps of the main label classifier and a confounder classifier. This approach decorrelates confounding features from latent representations during training. On the Spawrious Many‑to‑Many Hard Challenge benchmark, eX2L outperforms the current state‑of‑the‑art by 5.49% in average accuracy and 10.90% in worst‑group accuracy, while also demonstrating functional domain invariance through explicit label‑nuisance decoupling.
arXiv:2606. 18383v1 Announce Type: new Abstract: Sparse autoencoders (SAEs) are increasingly used to extract interpretable features from language models (LMs), yet a central question remains: when can an SAE-based explanation be treated as a faithful view of an underlying frozen LM We study this through a post-hoc generalization framework that certifies the LM via a sparse proxy, obtained by replacing a native hidden activation with its pretrained SAE reconstruction.
arXiv:2609.07876v1 Announce Type: cross Abstract: Recent methods in language model interpretability employ techniques such as sparse autoencoders to decompose residual stream contributions into linea...