You Only Pass Once: Answering and Abstaining Together in a Single Forward Pass of a Frozen Language Model

Ziyang Luo, Zhongyao Chu, Xinjie He, Youting Wang, Xukui Qin et al.|August 14, 2026arXiv

Key Takeaway

You can improve a frozen language model's reasoning and confidence detection in one forward pass by reconstructing clean internal states after steering, rather than running separate passes or accepting interference between techniques.

Summary

This paper solves a key problem in frozen language models: they both misuse their internal knowledge and fail to recognize when they lack sufficient information to answer.

reasoning efficiency evaluation

Key Terms

residual-stream steering-vectors abstention frozen-model sufficiency-detection