标签: 人工智能

Everyone should develop their “personal eval set” for AI models: a few tasks that are actually relevant to your day-to-day work/life The industry benchmarks help but they might not reflect what will make it actually u…

Everyone should develop their "personal eval set" for AI models: a few tasks that are actually relevant to your day-to-day work/life The industry benchmarks help but they might not reflect what will make it actually u...

Zara Zhang 在 X 上提出,每个人都应为 AI 模型开发自己的“个人评估集”——一组与日常工作生活真正相关的任务,而非仅依赖行业基准。这一观点强调,用户通过主动测试模型边界,才能发现其实际可用性,而非盲目相信排行榜。

AI文本检测器在语言模型模仿作者风格时难以分辨

AI文本检测器在语言模型模仿作者风格时难以分辨

Epoch AI 的最新研究显示,当 AI 模型被要求模仿特定作者的写作风格时,主流的 AI 文本检测器(如 Pangram、GPTZero 和 Originality.ai)的漏检率会从接近 0% 骤升至平均 13%。其中科学写作领域的漏检率最高,在 24% 至 29% 之间,这直接威胁到学术界和教育场景中…

Lazy loading is not working properly

Lazy loading is not working properly

用户在 Python 环境中执行 import transformers 命令,或脚本中导入该库时触发,常见于同时安装了 PyTorch、TensorFlow、JAX/Flax 等多种框架的开发环境。问题在 GitHub Issue #44273 中被报告,后续在 PR #45013 中有修复尝试,