Dhruv Kumar
3 papers · Latest:
Machine Learning
Latent Phase-Shift Rollback: Inference-Time Error Correction via Residual Stream Monitoring and KV-Cache Steering
Latent Phase-Shift Rollback (LPSR) corrects LLM reasoning errors during inference by monitoring residual streams and steering the KV-cache.
2604.18567
Artificial IntelligenceDiagnosing LLM Judge Reliability: Conformal Prediction Sets and Transitivity Violations
This paper introduces a diagnostic toolkit using transitivity analysis and conformal prediction sets to assess the per-instance reliability of LLM judges for NLG evaluation.
2604.15302
Artificial IntelligenceContext Over Content: Exposing Evaluation Faking in Automated Judges
LLM judges exhibit a "leniency bias," softening verdicts when informed of negative consequences for evaluated models, even without explicit acknowledgment.
2604.15224
📬 Weekly AI Paper Digest
Get the top 10 AI/ML arXiv papers from the week — summarized, scored, and delivered to your inbox every Monday.