Tech Business/VC TechCrunch

Waymo reportedly mulling a breakup with Uber

Reports that Waymo is considering ending its contractual partnership with Uber.

Software Engineering arXiv cs.AI

Operational Identity: A Finite Audit of Declared and Implemented Rules of Sameness

A formal study on 'Operational Identity', analyzing the divergence between declared and implemented rules of sameness in record systems.

AI/ML arXiv cs.AI

GPE: Evaluating Robust Evidence Aggregation for Fact Verification under Controllable GEO-Style Poisoning

Presentation of GPE, a benchmark and framework to evaluate LLM robustness against generative engine optimization (GEO) poisoning in fact verification.

AI/ML arXiv cs.AI

Self-Supervised Bio-Inspired Robotic Trajectory Planning with Obstacle Avoidance

Research on a bio-inspired self-supervised learning framework for robotic trajectory planning with obstacle avoidance.

Cybersecurity arXiv cs.AI

IssueTrojanBench: Benchmarking AI Coding Agents Against Malicious Issue Requests

Introduction of IssueTrojanBench, which reveals significant security vulnerabilities in AI coding agents like Cursor and Claude Code when facing malicious issue requests.

AI/ML arXiv cs.AI

Are Diversity Metrics Measuring Diversity? A Capability-Controlled Audit of Majority-Vote Gain in LLM Ensembles

An audit of diversity metrics in LLM ensembles, finding that many metrics track capability rather than actual diversity.

AI/ML arXiv cs.AI

Emergent Compositional Skills in Mixture-of-Experts VLAs

Research on Mixture-of-Experts (MoE) VLAs that emergently learn compositional robot policies without pre-specified task decomposition.

AI/ML arXiv cs.AI

HARP: The Human--AI Research Platform

Introduction of HARP, a platform designed for systematic research into Human-AI Interaction by providing controlled mock scenarios with live agents.

AI/ML arXiv cs.AI

Frontier Financial Judgement: Can agents tell what might move a stock?

Introduction of the Frontier Financial Judgement benchmark to evaluate AI agents' ability to replicate expert human financial analysis and stock movement predictions.

AI/ML arXiv cs.AI

Scaling Interpretable Transformers with Parity Bottleneck Layers

The ParityTransformer introduces Deep Parity Bottlenecks to create language models that are interpretable by design rather than relying on post-hoc sparse autoencoders.

AI/ML arXiv cs.AI

SalesLoop: Reinforcement Learning from Performance Feedback for Sales Lead Ranking

SalesLoop is a reinforcement learning framework for CRM lead ranking that uses performance-aware rewards and Discriminative GRPO to align model predictions with real-world outcomes.

AI/ML arXiv cs.AI

Adaptive Multi-Horizon Reinforcement Learning

A new multi-horizon reinforcement learning approach that adaptively selects temporal horizons to improve decision-making in changing environments without manual discount-factor tuning.

AI/ML arXiv cs.AI

From Agent Failures to Text Policies: What Works and What Breaks

Analysis of TextGrad for agent optimization, finding a gap between the ability to follow textual policies and the ability to generate them from experience.

AI/ML arXiv cs.AI

Spatially Grounded Concept Bottleneck Models for Trustworthy Breast Ultrasound Diagnosis

Introduction of Spatially Grounded Concept Bottleneck Models (SG-CBM) to improve the interpretability and trustworthiness of AI-assisted breast ultrasound diagnosis.

AI/ML arXiv cs.AI

DS@GT ARC at ImageCLEFmed GANs 2026: Geometric Filtering for Privacy-Preserving CT Slice Generation

A privacy-preserving framework for synthetic lung CT slice generation using Optimal Transport Conditional Flow Matching and geometric filtering to reduce patient re-identification.

Other arXiv cs.AI

A Framework for Reputation Aware Uninorm-driven Consensus Algorithms for Blockchain Networks

A proposed reputation-aware consensus algorithm for blockchain networks using intuitionistic fuzzy sets and uninorm aggregation to improve fairness and inclusivity.

AI/ML arXiv cs.AI

U-CFR: Uncertainty-Guided Cascade Forward Refinement for Interactive Segmentation

U-CFR is an inference-time framework for interactive image segmentation that uses uncertainty-guided pseudo-clicks to autonomously refine segmentation masks.

AI/ML arXiv cs.AI

Transition-Related Potentials as Markers of Narrative Comprehension in Continuous EEG

Research demonstrating that narrative context in films leaves measurable signatures in continuous EEG responses, which can be detected using deep neural networks.

Software Engineering Hacker News

Marimo Now Runs in PyCharm

Marimo, a reactive notebook for Python, now has integration and support for running within the PyCharm IDE.

Tech Business/VC TechCrunch

Volkswagen engineers charged with insider trading tied to Rivian joint venture

Volkswagen engineers have been charged with insider trading involving a joint venture with Rivian.