A nearly conserved 'energy' combining hidden-state velocity and next-token certainty is reported across LLMs, and a Jacobian steering method derived from it improves continuation quality.
Among RL-based models, POMO has demonstrated strong performance on a variety of tasks, in- cluding variants of the Vehicle Routing Prob- lem (VRP)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Momentum Point-Perplexity Mechanics in Large Language Models
A nearly conserved 'energy' combining hidden-state velocity and next-token certainty is reported across LLMs, and a Jacobian steering method derived from it improves continuation quality.