BLOG POST

There’s a quiet result from Cheng et al. (Renmin University and Microsoft Research, 2026) that I keep coming back to. They gave large language models access to a minimal sandbox — essentially a code interpreter with a file system — and watched what happened. No additional training. No new data.