#gradio

Why it matters: agents need reliable tools, not only larger base models. Clement Delangue put a concrete number on the shift, saying agents can now call 1M Hugging Face Spaces.

A Hacker News thread surfaced OBLITERATUS, an open-source project that studies and alters refusal behavior in open-weight LLMs without retraining. The interesting part is not just the capability claim but the project’s framing as a shared telemetry-backed research pipeline for comparing safety-editing methods across models and hardware.