一名大学生开发了一款应用程序，可以帮助我们破译文本是由人写的，还是由OpenAI疯狂的新聊天机器人ChatGPT生成的。普林斯顿大学计算机科学和新闻专业的学生爱德华·田(Edward Tian)说，他创建了这个程序，他称之为“GPTZero”，以帮助打击由新的人工智能聊天机器人产生的学术抄袭。
ChatGPT是OpenAI公司的新大型语言模型机器人，它能够吐出类似人类的文本，令观众惊叹不已。在Gizmodo，我们用这个程序做了很多事情，包括写一个完整的科幻小说故事，为我们写一篇博客。这项技术给很多人留下了深刻的印象，但也让他们感到担忧。批评者尤其担心，聊天机器人可能会毁掉大学论文，导致虚假信息激增，并在其他方面对主要媒体行业造成破坏。
因此，田的程序分析文本的复杂性和“随机性”，以评估它是由人类还是机器生成的——似乎是一件相当不错的事情。本周，这名大学生在推特上分享了他的创作链接，解释了它是如何“快速有效地检测出一篇文章是ChatGPT还是真人写作”。我在新年的时候创建了GPTZero，一个可以快速有效地检测文章是ChatGPT还是人类写作的应用程序。
GPTZero似乎工作得很好。在我第一次使用这款应用时，我插入了最近与ChatGPT对话的一些文本，几秒钟内，它就准确地推断出该文本是“机器生成的”。接下来，我从我最近的博客中输入了一些文章，同样，它很快就发现这是由人类写的。你在程序中插入的文本越多，结果看起来就越好。因此，如果你添加至少几个段落的文本，以获得准确的阅读效果，这是有帮助的。
GPTZero非常酷——尽管它只是表明，在我们的反乌托邦时代，机器不仅为我们写东西，而且还告诉我们它们是否写了东西。人类在未来会保持任何认知能力吗?还是机器人会为我们做所有的思考?对于我们未来的智商来说，情况似乎相当严峻。
原文：
A college student has created an app to help us decipher whether text was written by a human or generated by OpenAI’s crazy new chatbot, ChatGPT.
Edward Tian, a computer science and journalism student at Princeton, says he created the program, which he dubs “GPTZero,” to help combat academic plagiarism generated by the new AI-powered chatbot.
ChatGPT, OpenAI’s new large language model bot, has been stunning audiences with its ability to spit out human-like text. Here at Gizmodo, we have used the program to do a number of things, including pen an entire science fiction story and write one of our blogs for us. The tech has impressed a lot of people—but it has also worried them. In particular, critics fear that the chatbot will potentially doom the college essay, lead to a swell in disinformation, and prove otherwise disruptive to major media industries.
Thus, Tian’s program—which analyzes text for complexity and “randomness” to assess whether it was spawned by a human or machine—seems like a pretty good thing.
The college student shared links to his creation on Twitter this week, explaining how it was designed to “quickly and efficiently detect whether an essay is ChatGPT or human written.”
GPTZero seems to work pretty well. In my initial run with the app, I plugged in some text from a recent conversation with ChatGPT and, within seconds, it accurately deduced that the copy was “machine generated.” Next, I plugged in some writing from a recent blog of mine, and, again, it quickly figured out that it was written by a human. The more text you plug into the program, the better the results seem to be—so it helps if you add at least several paragraphs of copy for an accurate readout.
If you’re curious about how the whole thing works, you can head to Tian’s website to check it out for yourself.
GPTZero is pretty cool—though it just goes to show that, in our dystopian present, not only are machines writing stuff for us, but they’re also telling us whether they wrote it or not. Will humans maintain any cognitive abilities in the future or are robots going to do literally all of our thinking for us? Things seem pretty grim for our future IQs.
本文由数字化转型网（www.szhzxw.cn）翻译而成，作者：Sharon Machlis；翻译：数字化转型网郑亚茹；翻译审核：数字化转型网默然。
