2168
6How can you recognize AI generated text?
For most 18 to 24 year olds, generative AI is already part of daily life. Research, emails, translations, summaries, AI is everywhere. Its possibilities feel endless, limited only by your imagination.
But with that comes a growing skepticism, especially in schools and workplaces. Was this really written by a person? Or did ChatGPT write this assignment? The good news is, there are clues, they can tip you off.
What are these clues?
The project Voltaire blog points out that AI often uses vague generic language, where a person might say something is important, AI might say it's significant or crucial. Because chatbots know a little about everything but aren't true experts, they tend to stay broad and impersonal.
Another giveaway is the style. AI loves smooth, evenly paced sentences, joined together with lots of connectors. However, indeed, consequently. The tone stays flat, polite, and carefully neutral, instead of calling something a problem, it will soften it to a challenge.
But are there tools to detect generated text?
Yes, quite a few. Tools like GPT Zero and ZeroGPT, Draft, Goal, and Turnitin are designed especially for this. Even platforms like Scribber, which is for proofreading, now have AI detectors built in. They're simple to use, paste in the text, and the tool gives you a percentage. This looks 70% AI generated, for example. Some even highlight the suspected sentences. Just keep in mind, they're not flawless, false positives, and false negatives are common.
But how do these tools work?
They fight fire with fire using AI to catch AI, trained on huge databases of both human and machine written text, detectors learn to spot the telltale patterns, uniform sentence lengths, limited variation, and predictable structures.
They measure things like complexity, since simpler text often suggest AI, and diversity, because humans naturally vary sentence, rhythm, and style, while AI keeps it steady. Based on that analysis, they give you a probability score.
So, if you are using AI for writing, a quick fix is to vary your vocabulary and sentence lengths. And yes, you can run it through a detector, but remember, the tech isn't foolproof, at least not yet. There you have it.
词汇表
skepticism [ˈskeptɪsɪzəm] n. 怀疑态度，怀疑主义，怀疑论
tip someone off 向某人透露消息，给某人通风报信，给某人提示
vague [veɪɡ] adj. 含糊的，不明确的
generic [dʒəˈnerɪk] adj. 笼统的，通用的，一般的
impersonal [ɪmˈpɜːsənl] adj. 缺乏人情味的，客观的，非个人的，不带个人感情的
giveaway [ˈɡɪvəweɪ] n. 破绽，暴露真相的事物，泄露秘密的迹象
style n. 风格，文体
evenly paced [ˈiːvnli peɪst] 节奏均匀的，速度平稳的
connector [kəˈnektə(r)] n. 连接词，连接物
soften [ˈsɒfn] vt. 使缓和，使柔和，减轻
proofread [ˈpruːfriːd] vt. 校对，校阅，校读
detector [dɪˈtektə(r)] n. 检测器，探测器
built in 内置的，嵌入式的，内建的
flawless [ˈflɔːləs] adj. 完美无瑕的，无缺陷的
false positives 误报，假阳性
false negatives 漏报，假阴性
fight fire with fire 以毒攻毒，以其人之道还治其人之身
database [ˈdeɪtəbeɪs] n. 数据库，资料库
telltale [ˈtelteɪl] adj. 泄露秘密的，暴露真相的，明显的
variation [ˌveəriˈeɪʃn] n. 变化，变动，多样性
probability [ˌprɒbəˈbɪləti] n. 可能性，概率
quick fix 快速补救办法，权宜之计
foolproof [ˈfuːlpruːf] adj. 万无一失的，傻瓜式的
🪴翻译和pdf见公众号【琐简英语】，回复"1"可进入【打卡交流群】
