Tech Business/VC TechCrunch

Waymo and Uber quietly part ways in Phoenix

Waymo and Uber have ended their three-year partnership in Phoenix, Arizona.

AI/ML arXiv cs.AI

Chronic Kidney Disease Prognosis Prediction Using Transformer

Researchers introduced ProQ-BERT, a transformer-based framework for predicting Chronic Kidney Disease progression using multi-modal electronic health records.

AI/ML arXiv cs.AI

Towards Benign Memory Forgetting for Selective Multimodal Large Language Model Unlearning

The SMFA framework and S-MLLMUn Bench are proposed to enable selective unlearning of privacy-sensitive information in multimodal LLMs without degrading general capabilities.

AI/ML arXiv cs.AI

Health-ORSC-Bench: A Benchmark for Measuring Over-Refusal and Safety Completion in Health Context

Health-ORSC-Bench is a new large-scale benchmark to measure over-refusal and safe completion quality of LLMs in healthcare contexts.

AI/ML arXiv cs.AI

GAIA: A Data Flywheel System for Training GUI Test-Time Scaling Critic Models

The GAIA system introduces a data flywheel to train intuitive critic models that improve the test-time scaling performance of GUI agents.

Cybersecurity arXiv cs.AI

Just Ask: Curious Code Agents Reveal System Prompts in Frontier LLMs

The JustAsk framework demonstrates that autonomous code agents can be used to systematically recover hidden system prompts from frontier LLMs.

AI/ML arXiv cs.AI

CausalFlip: A Benchmark for LLM Causal Judgment Beyond Semantic Matching

CausalFlip is a new benchmark designed to test whether LLMs are performing true causal reasoning or merely relying on semantic matching.

AI/ML arXiv cs.AI

Conservative Equilibrium Discovery in Offline Game-Theoretic Multiagent Reinforcement Learning

COffeE-PSRO is introduced as a conservative equilibrium discovery method for offline game-theoretic multiagent reinforcement learning.

AI/ML arXiv cs.AI

SEA-TS: Self-Evolving Agent for Autonomous Code Generation of Time Series Forecasting Algorithms

SEATS is a self-evolving agent framework that autonomously generates and optimizes code for time series forecasting algorithms.

AI/ML arXiv cs.AI

Algorithms for Deciding the Safety of States in Fully Observable Non-deterministic Problems: Technical Report

A new policy-iteration algorithm, iPI, is presented to decide the safety of states in non-deterministic problems with guaranteed polynomial worst-case runtime.

Hardware/Chips Hacker News

Wallace the 6 inch f/2.8 telescope, building it, and hiking with it

A detailed account of building and using a custom 6-inch f/2.8 telescope, focusing on the process and practical application.

Other Hacker News

The Radiation Exposure Lie

A discussion questioning the common narratives around radiation exposure.

AI/ML Hacker News

Ornith-1.0: self-improving open-source models for agentic coding

Introduction of Ornith-1.0, a set of open-source models designed for self-improving agentic coding.

Tech Business/VC TechCrunch

Anthropic and Gov. Newsom forge deal allowing California government to use Claude at half price

Anthropic has reached an agreement with the California government to provide Claude AI services at a discounted rate.

Hardware/Chips TechCrunch

South Korean tech giants commit over $550B to ease ‘ RAMageddon’

South Korean memory chip giants commit $550 billion to expand fabrication capacity to address AI-driven RAM shortages.

Hardware/Chips The Verge

At $499, Apple’s M3-powered iPad Air is a good deal

A review and deal highlight for the M3-powered iPad Air, noting its performance gains and value at a discounted price.

AI/ML arXiv cs.AI

"Generate" the Future of Work through AI: Empirical Evidence from Online Labor Markets

An empirical study on how generative AI is displacing labor and inducing skill transitions in online programming-intensive job markets.

AI/ML arXiv cs.AI

Agentic Episodic Control

Presents Agentic Episodic Control (AEC), a new architecture integrating LLMs into reinforcement learning to improve data efficiency and generalization.

AI/ML arXiv cs.AI

Symmetry-Aware Transformer Training for Automated Planning

Introduces a symmetry-aware contrastive learning objective for Transformers to improve their efficiency in automated planning tasks.

AI/ML arXiv cs.AI

PreferThinker: Reasoning-based Personalized Image Preference Assessment

Proposes PreferThinker, a reasoning-based framework for personalized image preference assessment using a predict-then-assess paradigm.