AI's recursive self-improvement may not arrive as quickly as the industry claims. A new study found AI agents still can't conduct open-ended AI research, which requires judgment and creativity for genuine breakthroughs. Researchers question how crucial such research is to recursive self-improvement and whether AI systems can progress by improving narrower tasks.

This summer has been exceptionally hot across the Northern Hemisphere. June and July marked the hottest two-month stretch in Europe since record-keeping began, the contiguous US experienced its hottest July on record, and South Korea recorded its highest-ever temperature. Climate change intensifies heat waves, but El Niño is also amplifying global temperatures, with predictions that 2027 could be even hotter.

Open AI has paused some model work over safety concerns, citing that its Astra model reached a "critical" risk threshold. The company has made security updates following the Hugging Face hack and introduced a version of ChatGPT for teenagers. Meanwhile, Chinese humanoid maker Unitree surged 629% in its stock market debut, underscoring China's lead in humanoid robotics. China is also allowing Nvidia's H200 chips into the mainland, with ByteDense and Tencent each receiving about 10,000 processors.

Other notable stories include Flock's new AI tool enabling police to identify drivers by movement patterns, prosecutors arguing Meta deliberately hooked children on Facebook and Instagram, and ethical concerns surrounding Meta's smart glasses. Additionally, a blind Egyptian developer has built an AI app that helps others "see" by using cameras to answer questions about users' surroundings.