Read selected preprints, examine their limits, and follow the original research feeds.
Selected recent preprints · checked October 22, 2026. These notes summarize abstracts, not independent replications or peer-review conclusions. Follow the live indexes for newer submissions.
Sudipta Paul, Pengrui Quan et al.
Can an agent improve its planning harness with fewer evaluations?
Tracks the ancestry of harness edits and directs new edits toward diagnosed planning errors. The abstract reports better planning accuracy and lower evaluation budgets.
Harness evolution around frozen models; this is not evidence of autonomous frontier-model self-improvement. Read the experiments before generalizing.
Tianyou Lai, Wentao Yue, Zhiwei Zuo & Qilei Li
Do decision models follow rules consistently when wording and information change?
Uses executable rules and controlled changes to separate transfer, joint reliability and information use. Aggregate accuracy alone can conceal different failure modes.
A controlled decision-task study, not a universal product ranking. The abstract does not establish which commercial product is best.
Latest NLP and language-model papers
Latest machine-learning research papers
Daily-curated trending papers with community discussion
Discuss and annotate arXiv papers openly
Clear interactive research explanations (archived, still essential)
Paper deep dives on video