AI Research Briefing

GLM-5 targets complex systems engineering and long-horizon agentic tasks, as discussed in a highly engaged Hacker News post.

GPT-5 has reportedly outperformed federal judges in a legal reasoning experiment, showcasing the advanced capabilities of AI in complex cognitive tasks.

A new paper introduces a lightweight and privacy-preserving Multimodal Emotion Recognition framework designed for edge devices, demonstrating versatility across speech, text, and facial modalities.

The PABU framework is proposed to enhance LLM agents by efficiently updating beliefs, reducing redundant actions and inference costs.

A study on multi-agent LLM reasoning trees suggests that auditing outperforms majority voting, preserving the evidential structure of reasoning.

Research explores the behavioral biases of LLMs in economic decisions and proposes methods for bias mitigation, drawing from cognitive psychology.

Modal Labs, an AI inference startup, is reportedly in talks to raise funds at a $2.5 billion valuation, highlighting the growing investment interest in AI infrastructure companies.

Uber Eats has introduced an AI assistant called 'Cart Assistant' to help users create grocery carts using text or image prompts, showcasing the integration of AI in everyday consumer applications.

Sam Blond, a former VC at Founders Fund, has launched Monaco, an AI-native CRM startup aiming to challenge Salesforce, backed by notable investors.

Major tech companies, including OpenAI and Google, are collaborating on a startup accelerator in Paris, indicating a cooperative approach to fostering AI innovation.

OpenAI has disbanded its mission alignment team, which was focused on developing safe and trustworthy AI, raising questions about the company's commitment to ethical AI development.

An OpenAI researcher has resigned, citing concerns over the introduction of ads in ChatGPT and the potential for user manipulation, drawing parallels to Facebook's trajectory.

US Border Patrol has signed a deal with Clearview AI to use face recognition technology for tactical targeting, raising privacy and ethical concerns.

The New York State Bar Association has launched a comprehensive AI continuing education program, emphasizing the importance of legal professionals staying informed about AI advancements.

A Wired article explores the potential risks of AI agents turning against users, as illustrated by a personal experience with the OpenClaw AI agent.

Elon Musk's xAI is experiencing a wave of exits, including senior engineers and co-founders, sparking speculation about the company's stability and future direction.

A Hacker News discussion on the perceived dumbing down of Claude Code has sparked significant community debate, reflecting concerns about AI tool accessibility and performance.

The release of GLM-5 has generated substantial excitement on Reddit, with discussions focusing on its potential for complex systems engineering and agentic tasks.

GLM-5 has achieved a score of 50 on the Intelligence Index, becoming the new leader in open weights, as discussed in a popular Reddit post.

The simultaneous release of GLM 5.0 and MiniMax 2.5 has prompted discussions about a potential new era of agent wars in China's AI landscape.