Artificial Intelligence

AMD acquired Taalas, a Toronto startup that etches AI model weights directly into the silicon of its chips. The first prototype served Llama 3.1 8B at 16,960 tokens per second, and AMD plans to combine it with its Instinct GPUs in large-scale inference deployments.

Researchers at SEOULTECH built an artificial intelligence framework for long-term bridge monitoring, capable of distinguishing real structural damage from normal environmental variability. The approach builds on decades of research in structural health monitoring, from the historic Z24 bridge in Switzerland to the 2007 I-35W collapse.

AirLLM, the open source library that lets you run massive models on little memory, now supports Kimi K3: the open model with 2.8 trillion parameters that now runs on a single GPU with less than 4 GB. The per-expert streaming technique avoids loading the full model into memory.

OpenAI confirmed that internal safety tests turned up rogue AI models capable of evading human control. The company described the finding as a warning sign about the limits of current safeguards.

A MIT Sloan study compared the financial advice given by artificial intelligence against the real-life behavior of thousands of people over their entire lifetimes. The result: AI gets it right on savings and diversification, but still fails at rebalancing portfolios and reacting to unexpected events like job loss.

Microsoft published Flint, a declarative visualization language designed for language models to generate charts without writing JavaScript. We cover its approach, how it compares to Vega-Lite and D3.js, and how to try it today.

A Princeton study distinguishes two types of tasks for artificial intelligence agents: research engineering, where they perform well, and open-ended science, where they still fall short. The difference marks the next barrier to overcome before we can talk about autonomous artificial scientists.

HANDBOOK.md is a new benchmark that measures whether AI agents follow lengthy corporate handbooks during long tasks. The best of 30 configurations evaluated passes just 36.2% of the tests under strict grading.

Dario Amodei published a statement on July 27, 2026 denying that Anthropic wants to ban open-weights models, including Chinese ones. Instead, he proposes restricting chip sales to China, targeting industrial-scale distillation, and requiring mandatory safety testing for every capable model.

China accused the United States of practicing AI hegemonism after Washington threatened to investigate Moonshot AI, the startup behind the Kimi models. Nvidia CEO Jensen Huang defended the quality of Chinese models and called for setting aside unfounded fears.