Showing 2026Show all
2 papers · 1 filter
cs.CL2026
Prompt Compression via Activation Aggregation
Thibaud Ardoin, Semira Einsele, Evis Bregu +1
Large language models process prompts by propagating activations through dozens of layers before generating a response. We ask whether the task-relevant information contained in an…
cs.AI2026
LLM Self-Recognition: Steering and Retrieving Activation Signatures
Thibaud Ardoin, Jonas Schäfer, Gerhard Wunder
Recent advances in interpretability suggest that large language models (LLMs) implicitly encode signals in their generated text that enable self-recognition of their outputs. We de…