Belief-Trajectory Energy: One Forward Pass Yields Difficulty, Review Detection and Attribution
Jiahao Ying · hf · 2026-10-19
The paper introduces Belief-Trajectory Energy (BTE): one forward pass per token measures how input reshapes layerwise predictions in a language model, stacking JSD between consecutive readout distributions into a training-free scalar. Signals support multiple uses: estimating task difficulty, detecting AI-generated reviews via logistic regression on BTE features, and generator attribution via softmax(W·e) across eight sources. A cheap new lens on a model's internal belief path.
More from Research
- Stanford and Databricks use AI agents to reproduce medical findings and test analytical robustness — ChengleiSi · 2026-10-23
- EMNLP paper: diffusion LMs internally encode latent denoising timestep representations — gsarti_ · 2026-10-23
- KU Leuven and VIB open two Group Leader professorships in AI for Biology, start 2027 — jpillowtime · 2026-10-23
- Subspace Tuning curbs LLM catastrophic forgetting by freezing principal singular directions — scottniekum · 2026-10-23
- Perceptic launches PercepticOS, an AI workspace for 'scientific superintelligence' — nathanbenaich · 2026-10-23
- EPFL's Bosselut to talk long context as test-time learning at ELLIS seminar — andre_t_martins · 2026-10-23