I Asked Two Local LLMs to Operate Linux. The Score Hid the Dangerous Failures.
中文摘要
研究显示 Qwen3 与 Nemotron 4B 在 Linux 任务中的高分可能掩盖了危险的操作失败。
English Summary
Testing Qwen3 and Nemotron 4B shows that high scores in Linux tasks can mask dangerous operational failures.
原文节选
300 reproducible responses from Qwen3 4B and NVIDIA Nemotron 3 Nano 4B on SSH, Linux, Docker and DevOps safety Continue reading on Medium »