Deft 发布:AI 检测工具易被愚弄,写作工具应引导正确模式

elvis · @omarsar0 · X·2026-08-19 09:59·17天前
AI 导读

Deft 联合创始人 @rosmine 宣布推出新 AI 写作实验室 Deft,其模型生成的文本据称 86% 被 pangram 判定为“完全人类”。Elvis Saravia 认为 AI 检测工具极易被规则欺骗、纯属浪费时间,主张开发 Deft 这类工具以鼓励负责任的写作模式,而非依赖检测。Deft 目前为小规模测试版,在分析/论文、创意写作和改写场景表现较好。

elvis@omarsar0
37AI 编辑部评分,满分 100

Deft 发布:AI 检测工具易被愚弄,写作工具应引导正确模式

2026-08-19 09:59· 17天前
AI 导读

Deft 联合创始人 @rosmine 宣布推出新 AI 写作实验室 Deft,其模型生成的文本据称 86% 被 pangram 判定为“完全人类”。Elvis Saravia 认为 AI 检测工具极易被规则欺骗、纯属浪费时间,主张开发 Deft 这类工具以鼓励负责任的写作模式,而非依赖检测。Deft 目前为小规模测试版,在分析/论文、创意写作和改写场景表现较好。

Great initiative. We don’t need to ignore the use of AI in writing, we just need better tools that encourage the right type of writing patterns and avoid “slop.”

BTW, I don’t trust any of these AI detection tools. I asked Fable to build a simple list of rules that trick all of these tools into thinking a human wrote a piece of text. It’s really easy. It’s baffling to me that so many smart people here think these AI detection tools work. They simply don’t. I didn’t do this because I want to trick the detection tool but to show how easy it is to fool these systems. AI detection tools are a waste of time and counter productive. Instead, I believe we should try to build AI tools like Deft to help with more responsible writing that’s free of all the undesired patterns.

RosmineAnnouncing Deft, a new AI lab for better writing, cofounded with @jmrphy See the picture for launch announcement the Deft model wrote for itself Currently, 86% ...

来源:elvis· x.com