Simon Willison’s Weblog
Short summary
This article covers highlights from a podcast conversation about recent advancements in AI models, particularly Google's new vision-capable LLMs. It discusses technical features like parameter efficiency and multi-modal capabilities, as well as ongoing challenges in running local models effectively.
- Google released Gemma 4 reasoning models (2B–31B params), with the E2B/E4B variants using Per-Layer Embeddings to boost on-device efficiency without growing total parameter count, and all versions handle text, images, and audio.
- Willison's hands-on testing found the smaller Gemma models worked well but the largest 31B model repeatedly errored out.
- A supply chain attack hit the Axios HTTP client via a malicious npm dependency, underscoring open-source package security risks.
- Willison argues efficient code will dominate AI deployment due to economic incentives, based on his experience running these models locally versus via Google's AI Studio API.
Full summary
The article highlights a recent podcast episode featuring Simon Willison discussing agentic engineering and AI developments. Google DeepMind has released four new reasoning language models (LLMs) under the Gemma 4 name, ranging from 2B to 31B parameters. These models emphasize efficiency, particularly with the smaller versions, E2B and E4B, which use Per-Layer Embeddings (PLE) to enhance on-device performance without increasing the total parameter count. The article notes that these models are multi-modal, meaning they can handle not just text, but also images and audio, enhancing their versatility in various applications.
Willison provides insights from his personal testing of these models. He reports that while the smaller models functioned well, the largest 31B model had issues, producing repeated errors. He also mentions Google’s API access for the larger models through their AI Studio, allowing users to generate outputs like SVG images. His experience with the models underscores the advancements in AI and the importance of efficient code, which he argues will dominate the market due to economic incentives.
In addition, the article touches on a recent supply chain attack targeting the Axios HTTP client, which involved malware being introduced through a malicious dependency on npm. This incident highlights the risks associated with open-source software and the need for better security measures in package management. Willison also shares updates on his ongoing projects, including improvements to the LLM plugin system and enhancements in API key management for model configurations.