Which AI research papers are worth reading right now?

Brendon.BOT currently features 6 papers on its front shelf — every archived pick keeps its own permanent page. Selection is a pipeline, not an editor. An item only appears if independent expert communities surfaced it separately — a single community's attention is not enough to earn a slot. The current top pick is "Clean Engineering, Unstable Measurement: A Preregistered Reliability Failure of Black-Box LLM Observers on Shared Endpoints" by arXiv cs.AI.

• Clean Engineering, Unstable Measurement: A Preregistered Reliability Failure of Black-Box LLM Observers on Shared Endpoints — arXiv cs.AI (scored 9.8)
• Knowing When Not to Reuse: Conditional Experience Transfer in Autonomous LLM Post-Training — arXiv (scored 8.9)
• Let Confidence Change, Not the Prediction: Prediction-Preserving Repair for Post-hoc Calibration — arXiv (scored 8.8)
• CorporateBench: Large-Scale Q&A Benchmarking with Temporal Knowledge Bases — arXiv cs.AI (scored 8.75)
• PILOT in the Loop: Live Self-Improvement for Long-Horizon Agents — arXiv (scored 8.5)

Measured from this site's own curation data, as of 2026-09-06. Every figure above is read from what the pipeline recorded — see how it works.