Applying STP at consecutive semantic reasoning steps achieves 168x more accurate multi-step latent prediction on ProcessBench than frozen baselines, with trajectories forming smooth curves best captured by non-linear predictors.
A statistical physics of language model reasoning
2 Pith papers cite this work. Polarity classification is still indexing.
2
Pith papers citing it
years
2026 2representative citing papers
A density-matrix order parameter exposes silent fine-tuning failures and shows that sharp 'catapult' jumps are softmax readout artifacts, not genuine phase transitions.
citing papers explorer
-
Semantic Step Prediction: Multi-Step Latent Forecasting in LLM Reasoning Trajectories via Step Sampling
Applying STP at consecutive semantic reasoning steps achieves 168x more accurate multi-step latent prediction on ProcessBench than frozen baselines, with trajectories forming smooth curves best captured by non-linear predictors.
-
Phantom Transitions in Language Model Fine-Tuning: A Density-Matrix Analysis
A density-matrix order parameter exposes silent fine-tuning failures and shows that sharp 'catapult' jumps are softmax readout artifacts, not genuine phase transitions.