AI News — Thursday, May 28, 2026

9
LocateAnything: Fast and High-Quality Vision-Language Grounding with Parallel Box Decoding

Researchers introduce LocateAnything, a novel vision-language grounding model that achieves fast and high-quality object localization through an efficient parallel box decoding mechanism.

Hugging Faceresearch
9
Snowflake Signs $6B Deal with AWS for AI CPU Chips, Boosting Amazon's Cloud Business

Snowflake has reportedly signed a massive $6 billion deal with AWS to secure AI CPU chips, signaling a significant investment in AI infrastructure and a win for Amazon's cloud services.

TechCrunchindustry
8
How Are Developers Actually Using AI At Work?

A new report explores the practical applications and integration of AI tools by developers in their daily work routines, highlighting current trends and challenges.

Dev.toindustry
8
EvalVerse: Pipeline-Aware and Expert-Calibrated Benchmarking for Professional Cinematic Video Generation

EvalVerse introduces a new benchmarking framework designed for professional cinematic video generation, offering pipeline-aware and expert-calibrated evaluations to assess model performance accurately.

Hugging Faceresearch
8
Cisco and OpenAI Redefine Enterprise Engineering with Codex

Cisco and OpenAI announce a collaboration to integrate Codex, OpenAI's advanced AI model, into enterprise engineering workflows, aiming to enhance productivity and innovation.

OpenAI Blogproduct
7
Warp's Big Bet on Building Open Source with GPT-5.5

Warp is making a significant commitment to open-source development by leveraging GPT-5.5, indicating a growing trend of advanced AI models powering collaborative software creation.

OpenAI Blogopen-source
7
SpatialBench: Is Your Spatial Foundation Model an All-Round Player?

SpatialBench introduces a comprehensive benchmark to evaluate the versatility and capabilities of spatial foundation models across various tasks and domains.

Hugging Faceresearch
7
Why Google's AI Can't Spell Google (or Anything Else)

This article delves into the persistent challenges Google's AI models face with basic spelling, exploring the underlying reasons and implications for AI literacy.

TechCrunchindustry
7
Building Self-Improving Tax Agents with Codex

OpenAI showcases a new application of Codex in developing self-improving AI agents specifically designed to handle complex tax-related tasks, demonstrating practical agentic AI capabilities.

OpenAI Blogproduct
6
Google I/O 2026: Catch Up on the Dialogues Stage

A recap of the Dialogues stage at Google I/O 2026 highlights key discussions and announcements related to AI, innovation, and developer tools.

Google AI Blogindustry
6
MobileGym: A Verifiable and Highly Parallel Simulation Platform for Mobile GUI Agent Research

MobileGym offers a new simulation platform that provides verifiable and highly parallel environments for advancing research into AI agents interacting with mobile graphical user interfaces.

Hugging Faceresearch
6
Personalize-then-Store: Benchmarking and Learning Personalized Memory for Long-horizon Agents

This research introduces a new approach and benchmark for developing and evaluating personalized memory systems crucial for long-horizon AI agents.

Hugging Faceresearch
6
D^2-Monitor: Dynamic Safety Monitoring for Diffusion LLMs via Hesitation-Aware Routing

D^2-Monitor proposes a dynamic safety monitoring system for Diffusion LLMs that uses hesitation-aware routing to enhance safety and reliability in generative AI applications.

Hugging Faceresearch
5
Share More, Search Less: Collaborative Parallel Thinking for Efficient Test-Time Scaling

This paper explores a method called 'Collaborative Parallel Thinking' to improve the efficiency of test-time scaling in AI models by encouraging agents to share information and reduce redundant searches.

Hugging Faceresearch
5
The Open/Closed Problem in AI

This blog post discusses the fundamental 'open/closed problem' in AI, examining the trade-offs between open-source and proprietary approaches in AI development and deployment.

Lobste.rsindustry