Probe-Guided Early Exit

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• Probe-guided early exit reduces tokens by up to 80% on MMLU and 30% on GPQA-Diamond with similar accuracyclaim0.778
  Practical efficiency claim for using activation probes to enable adaptive computation
• Probe-guided early exit reduces tokens by up to 30% on GPQA-Diamond with similar accuracy on DeepSeek-R1 671B and GPT-OSS 120Bfinding0.766
  Quantitative efficiency result on hard benchmark, smaller reduction reflecting genuine reasoning need
• Probesconcept0.740
  Interpretability tools that decode information from internal model activations; here, linear probes are used for data attribution.
• Probe Generalizationconcept0.715
  The ability of probes trained on one dataset to transfer accurately to topically and structurally different datasets
• Impulsivity probe (impulsive vs. planning)concept0.709
  One of four emotive concept probes trained; contrastive pair impulsive/planning with best layer 13 in LLaMA-3.2-3B
• Diagnostic Probingmethod0.708
  Earlier interpretability method applying classifiers to DNN hidden representations; shares complexity-accuracy dilemma with causal abstraction
• For simple factual tasks F0-F3, probe directions show a sharp geometric transition in middle layers, with late-layer probes converging to high cosine similarity; A3 and F4-F5 show no clear transition.finding0.700
  Geometric evidence for convergence to stable truth directions only for simpler tasks.
• Probing Methodsmethod0.699
  Top-down interpretability approach studying linguistic properties at various residual stream stages; contrasted with the paper's bottom-up mechanistic approach