Dave Rupert shares a nuanced take on small language models (SLMs) running natively in the browser via Chrome and Edge's experimental Prompt, Summarizer, and Rewriter APIs. He outlines genuine appeal: local/private execution, no API keys, no per-token costs, lower energy use, and a low barrier to prototyping. However, he also endorses Mozilla's concerns about model calcification (sites tuning prompts to Google's specific model, echoing the Chrome-only era), model neutrality (Google's TOS becoming the de-facto standard), and the false promise of universal access when GPU-equipped devices are required. He proposes alternatives including a bring-your-own-model marketplace, stronger fallback stories for low-end devices, and spec-level guidance on prohibited use cases — or abandoning the spec entirely if accuracy problems can't be solved.
Sort: