I tested GPT-5's coding skills, and it was so bad that I'm sticking with GPT-4o (for now)

ZDNet - Artificial Intelligence

Aug 11, 2025 03:50

1 views

aibusinessenterprisetechnology

Summary

A recent coding benchmark found GPT-5’s performance disappointing, producing buggy code and confidently incorrect answers that require close human supervision. This suggests GPT-5 may not yet surpass GPT-4o for coding tasks, highlighting the ongoing need for caution and oversight when using advanced AI models in software development.

In my latest coding benchmark, GPT-5 stumbled badly, delivering broken plugins, flawed scripts, and confidence-laden wrong answers that could derail projects without careful human oversight. Here's what to know before you use it.

Read Full Article More News

Improving Your LLM Agent with Reinforcement Learning

Hacker News - AIAug 11

Microsoft’s Agent Lightning project explores how reinforcement learning can enhance large language model (LLM) agents, enabling them to learn from interactions and improve task performance over time. This approach aims to make LLM agents more adaptable and effective in real-world applications, signaling a significant step forward in the development of autonomous AI systems.

Blending AI, Sustainability and Speed: Exclusive Insights Into On2Cook’s Culinary Revolution

Analytics InsightAug 11

On2Cook has unveiled a new cooking device that leverages AI to optimize cooking times and energy usage, blending sustainability with speed in the kitchen. By using AI-driven algorithms, the device can intelligently adjust cooking parameters, reducing energy consumption and food waste. This innovation highlights the growing role of AI in promoting sustainable practices and efficiency in everyday appliances.

Online news publishers face extinction-level event from Google AI-powered search