马里兰大学与 Google DeepMind 研究：仅凭结构即可 93% 准确区分 AI 与人类写作

以下是视频口播的自动转写，可能有错字。

Why is ai so bad at writing? Because this is what ai writing looks like, and this is what human writing looks like. And now researchers have found a way to tell them both apart without even reading a single word.

A team from the university of maryland and google deepmind tested this on over 61,000 stories, and the structure alone separated human writing from ai with a 93% success rate.

And look, let's be real. We all know that you're using ai to write everything from your emails to your documents, at least i am.

So please at least make sure it sounds like you. But don't worry, i got you. Here are three skills that will actually help you so you don't get caught out in your job or at school.

Welcome to episode 2 of the 60 second ai academy, where i teach you everything actually worth learning about ai in under a minute.

First is humanizer. It catches common ai writing patterns like inflated language, repetitive phrasing, and unnecessary explanations.

And if you give it a draft of a few examples of your own work, it knows what to tailor the ai writing towards to avoid the generic slop.

Speaking of slop, the second is stop slop. Please never. Tell ChatGPT or claude to sound more human ever again. Instead, give it this skill to give it specific phrases and structures of sentences to avoid.

Now, cleaning up your drafts only gets you so far. The last and most important is teaching claude how to actually write for you like you.

And that's where anthropic skill creator comes in. Give it all your best emails, posts, and messages, and then use it to build the skill around your own voice.