Read this if you use ChatGPT to cheat
- Caimile Loy
- Aug 7, 2024
- 2 min read
Updated: Aug 19, 2024
OpenAI has reportedly made an AI detection tool with 99.9% accuracy... But they're afraid to release it to the public.
Everyone is obsessed with AI—especially Avi Schiffmann, a Harvard student who dropped $1.8 million on a domain URL for his AI product, a necklace called Friend. Crazy, genius, or both?
I love quotes that give us a new perspective to consider. Take a look at this one:
“Remember, AI won’t take your job… Somebody who is better at using AI than you will.” -Shelly Palmer
Read this if you use ChatGPT to cheat
I know it, you know it, we all know it: AI detectors can often be unreliable. But what about one that’s created by one of the top AI companies themselves, OpenAI?
Side note: I found an AI detector named ZeroGPT and another AI detector named GPTZero. Just thought it was funny. I'm a sucker for giving things unique and random names over slick and basic ones.
See how Google's Gemini got its name.
Sam Altman and his crew at OpenAI are hesitant to give the green light on an AI detection tool that is said to have exceptionally high accuracy (99.9%, to be exact), the Wall Street Journal reported on Sunday.
According to reports by Gizmodo, the company has had this secret tool under its belt for two years. With a push of a button, they could release it to the public. So, why haven’t they?
Nearly one-third of ChatGPT enthusiasts admitted that the anti-cheating tool would make them think twice about using ChatGPT. Professors would be thrilled. OpenAI… maybe not as much.
The tool might not work equally for everyone, and people whose native language isn’t English would take on the brunt.
OpenAI employees have raised concerns that (true) cheaters would find loopholes around the tool, anyway.
AI Detection loopholes
🤖 When ChatGPT writes something, it can change how it picks the next word. These small changes create a pattern that shows who—or what—made it, forming a watermark that’s invisible to the human eye.
Find a way to erase the watermark, and you’ve scored a loophole around the AI detector. Here’s what employees are afraid of that could successfully outsmart the tool
Users could translate ChatGPT text into another language and then back into English.
Users might try prompting ChatGPT to include emojis in its text before manually removing those emojis themselves afterward.
Others? I'm sure there are plenty.
Everyone's got their opinions, but there's one thing OpenAI employees all nod their heads to: deciding who gets to use this tool is definitely going to be a challenge.
