Towards interpreting any residual stream direction by scaling activation inversion
Celeste De Schamphelaere, Juan J. Vazquez, Ari Spiesberger, Tomáš Gavenčiak, Gavin Leech, Euan Ong, Jiahai Feng
Abstract
To interpret a direction in a language model's residual stream, we can look for text that strongly activates it. But when such text is found by scanning a corpus, it is limited to what that corpus contains. We develop Maximally-Activating Example Metamodels (MAEMs), language models which map residual stream directions from a subject model to text exemplars that strongly activate those directions. Our objective is to maximise the cosine similarity of a supplied residual stream direction with the activations of the text generated by the MAEM. This label-free RL objective helps to scale MAEMs on diverse sources of residual stream directions. Empirically, we find the resultant metamodel produces informative exemplars: a Qwen3.6-27B MAEM inverts held-out activations better than corpus search or a natural-language autoencoder (NLA), beats corpus search (at 10M tokens) on a majority of held-out languages/domains, and activates 89% of held-out SAE features. It also identifies concepts behind steering vectors about as well as NLAs, recovers the payloads of LoRA backdoors more often than corpus search, and can even verbalise some model workspace inner monologue. Unlike explanations produced by other metamodels (e.g. NLAs), MAEM exemplars can be verified with a forward pass on the subject model, making the MAEM objective a grounded alternative for scaling the training of metamodels.
Create a lesson
Related papers
Bio-PM-Guided Synthetic IMU Generation
Deepali Singh, Deepesh Parpani, Aviral Gupta
Active Feature Acquisition by Distilling an Oracle Teacher
Yang Meng, Zixin Ding, Robert L. Grossman et al.
The Window Lever: Longer Observation Windows Outweigh the Observed Effects of Data, Capacity, and Architecture in Customer Behavior Forecasting
Tara Sowrirajan, Yanqiao Wang
Hybrid Retrieval for Government Scheme Search in English and Marathi
Aishwarya Gunjal, Kanchan Khemnar, Pawan Poojary et al.
Grid Cells as Spectral Successor Features: A Reusable Basis for Continual Reinforcement Learning
Élodie Côté-Gauthier, Doina Precup, Isabeau Prémont-Schwarz
A new algorithmic approach for finding flat minima: critically damped Polyak stepsizes
Francesco Caporali, Jason M. Altschuler, Boris Hanin