Weekly Digest
Week of October 26, 2026
7 articles published this week
Monday, October 26
Patronizing and Condescending Language Detection in English and MandarinNew research shows that fineâtuned BERT models can detect patronizing and condescending language in English and Mandarin with modest success, highlighting both the challenge and potential of automated bias detection.
End-to-End Differentiable Diffusion Conditioning for Physics-Informed OptimizationDifferentiable Diffusion Conditioning (D2C) turns a frozen conditional diffusion model into a gradientâbased optimizer for offline modelâbased optimization, enabling physicsâinformed design improvements without testâtime access to the true objective.
Towards Adversarially Robust Multigraph Neural Networks for Financial Crime AnalysisMultigraph neural networks improve AML detection but are vulnerable to learned structural attacks, and adversarial training can bolster their robustness.
Don't Forget What You Just Learned: Self-Directed Repair Turns Agent Mistakes Into Lasting LessonsSelfâDirected Repair lets AI agents turn mistakes into permanent, lightweight adapters that boost future performance without costly fineâtuning.
PaperFormalizer: Author-in-the-Loop Autoformalization of Research Papers in LeanPaperFormalizer is an authorâinâtheâloop system that converts LaTeX papers into Leanâverified code, letting researchers close proof gaps while retaining full control over interpretation.
Phronix: The Capability-Centric Learning Paradigm for AgentsPhronix introduces a predictorâvalidatorâcorrector loop that lets AI agents learn reusable capabilities from their own executions, dramatically improving task success and efficiency.
Tradeoff-Aware Territory Design: A Multi-Objective Decision System for Enterprise Customer-to-Seller AllocationA multiâobjective decision system combines predictive account scoring, continuity cost estimation, and optimization to allocate enterprise sellers to customers, letting planners balance growth, balance, capacity, and relationship continuity.
Free weekly email
This week in AI
Every Monday, the weekâs AI briefings in one short email: what happened, why it matters, and the source. Unsubscribe anytime.