Back to feed
arXiv cs.LG
arXiv cs.LG
7/15/2026
Mirror Horizon: Viable Path Entropy as a Measure of Bounded Reflection

Mirror Horizon: Viable Path Entropy as a Measure of Bounded Reflection

Short summary

This paper formalizes Mirror Theory's viable path entropy (VPE), a measure of an AI system's verified continuation capacity under bounded reflection. VPE decomposes capability into reachability of viable continuations and diversity of verified modes. Experiments on Qwen2.5 models over GSM8K show that increasing token budget from 96 to 160 substantially expands verified reachability, with Qwen2.5-1.5B outperforming the larger 3B model at 160 tokens, demonstrating that capability is not just parameter count.

  • •Formalizes viable path entropy as a measure of LLM reasoning capability under bounded reflection
  • •Decomposes capability into reachability probability and diversity of verified continuation modes
  • •Qwen2.5-1.5B outperforms 3B model at 160-token budget, showing capability ≠ parameter count

Generated with AI, which can make mistakes.

#research-breakthrough
Read full article at arXiv cs.LG

Is this a good recommendation for you?

Comments

Failed to load comments. Please try again.

Explore more

When Does In-Context Search Help? A Sampling-Complexity Theory of Reflection-Driven Reasoning
arXiv CS.AI
Position: Quantum Program Generation Must Prioritize Validity Over Probabilistic Scaling
arXiv cs.LG
Cost-Effective Agent Harnesses for Abstract Reasoning and Generalization on ARC-AGI-1
arXiv CS.AI
Capability from Access Structure, Not Scale: Lower Bounds and Pre-Registered Tests for Hybrid Sequence Models
arXiv CS.AI