According to a survey by BestColleges, More than half of students use AI to cheat. Those numbers are consistent with a Stanford University study that found 60 to 70 percent of students cheat. However, AI may soon no longer be the lazy student’s answer to writing an essay. A Wall Street Journal (WSJ) article reported that “OpenAI has a reliable method detect when someone uses ChatGPT to write an essay or research paper” — with 99.9% accuracy.
Also: How AI Lies, Cheats, and Flatters to Succeed — and What We Should Do About It
As my colleague David Gewritz pointed out, Many programs have promised to detect AI-written text.However, he concluded, “I certainly would not feel comfortable threatening a student’s academic standing or accusing them of cheating based on the results of these tools.”
OpenAI hasn’t revealed details about how their new method can be nearly perfect at recognizing AI-written text. It certainly isn’t because it can detect AI hallucinations. It can’t. As OpenAI co-founder John Schulman said last year, “Our biggest concern is about realism because Models love to make things.“
That will never change. According to Mohamed Elgendy, co-founder and CEO of Kolenaa machine learning testing service,” the rate of hallucination will decreaseBut it will never go away — even highly educated people can give out misinformation.”
Rather than using some magical method to detect AI text, it appears that OpenAI is using a much simpler way to recognize AI-written text: The service may be copyrighting its results.
In a newly revised blog post, Understanding the origins of what we see and hear onlineOpenAI revealed that it has been working on using classifiers, watermarks, and metadata to detect AI-generated products. We still don’t know exactly how this watermark works.
Also: The Best Free AI Courses (and Are AI Certifications Worth It)
We know that OpenAI reports that it is “highly accurate and even effective against localized forgeries, such as rewording.” However, watermarking is “less effective against global forgeries.”
That means it doesn’t work well on translated text or something as simple as inserting special characters into text and then removing them. And of course, it can’t detect works from another AI model. For example, if you feed the ChatAPT AI text detector a document generated by Google Gemini or ConfusionIt probably won’t be able to identify it as an AI-generated document.
In short, with a little extra effort, students and writers can still claim AI chatbot work as their own. Yes, they can try. In my experience with AI, the results still tend to be second-rate. But if it’s good enough to get you a passing grade, it might be all you need.
At least one self-proclaimed professor on Reddit wasn’t impressed: “The thing is, you can copy-paste the text into another program, translate it into another language, and then translate it back. But honestly, most students won’t do that, so It can catch almost anyone.“
Also: How Pearson’s AI Assistant Can Save Teachers Time
Of course, that might not bother OpenAI CEO Sam Altman, who told The Harvard Gazette, “Cheating on homework is obviously bad.. But the meaning of cheating and the expected rules change over time.”
I don’t know about that. Cheating is cheating, but this new tool in OpenAI’s arsenal doesn’t seem to do much to prevent it.
Curiously, while OpenAI is still wrestling with when — or whether — to release this new service, the company will soon release a DALL·E 3 source classification. This means that, ultimately, almost every image you create with DALL-E will be marked as an innovative product of DALL-E AI. OpenAI relies on C2PA metadata, a digital content standard, for labeling and identifying images. If you’re a graphic designer who relies on DALL-E to create “native” graphics, it might be time to turn back to Photoshop.
