Super grateful to have received senior area chair highlight at #ACL2025NLP.⏳ The generalization of interpretability-based steering methods is at an inflection point.🚂 As a community, we need to place strong emphasis on methods-reliability evals if we care about long-term impact.

🌟Excited to announce that “Steering off Course” was accepted to #ACL2025NLP for an Oral and Panel Discussion! 📍Wed, 9AM, Level 2 Hall A. 🇨🇦 I will also share this work at Actionable Interpretability @ActInterp at #ICML2025.📍Sat, 1PM, East Ballroom A.

RT @HananeNMoussa: 📢Excited to announce the first project of my PhD!. Through our work we address the training data scarcity to develop AI….

We hope to inspire further research into internal transformer mechanism variance, so future steering methods can be robust and adaptable to new releases 🥕🐇💨. Special shoutout to my advisor @shocheen and collaborators @HariSethuraman @dheerajgopal @HannaHajishirzi.(10/10).

We report many aggregated results in our paper, and invite researchers to comb through the extensive results in our repository to build intuitions about model variance. Our paper: Code, Data, Results, and Figures for all LMs: (9/10).

✨Localization hypothesis does not always hold for FVs✨. Simple word-pair ICL translation from English into another language require more heads to be effective. Post-trained models are more steerable when more heads are used in the FV.(8/10)

✨Localization hypothesis does not always hold for FVs✨. FVs rely on the assumption that information required for ICL is stored and activated within a small subset of heads 🎯. But certain models require many heads in their FV before recovering performance 🙉🙉🙉➡️📈.(7/10)

🔔Neither Function Vectors nor Task Vectors are generalizable🔔. Several model families, even after significant hyperparameter tuning, show no improvement or even decline in relevant steering metrics🧰📉.(6/10)

🔔Neither Function Vectors nor Task Vectors are generalizable🔔. Even with the best-performing tool, FVs with full hyperparameter search, only 76% of model-task combinations recover 50% of 5-shot performance.(5/10)

DoLa’s poor efficacy could stem from the flawed assumption that factual.knowledge of an LM evolves gradually across layers. Correct/incorrect answer tokens have low probabilities before spiking at the same layer, suggesting contrasts with early layers are uninformative.(4/10)

DoLa contrasts token probabilities across layers to enhance factuality. ✅Consistent with prior work, we find that DoLa works decently for Llama 1 TruthfulQA and FACTOR. ❌However, for all other models tested, the improvements afforded by DoLa in most metrics is negligible.(3/10)

We examine steering methods inspired from Logit Lens and Activation Patching, specifically:.1️⃣DoLa 2️⃣Function Vectors 3️⃣Task Vectors (2/10).

Steering language models by directly intervening on internal activations is appealing–but does it generalize?. We study 3 popular steering methods with 36 models from 14 families (1.5-70B), exposing brittle performance and fundamental flaws in underlying assumptions.🧵👇.(1/10)