floofloof‘ChatGPT detector’ catches AI-generated papers with unprecedented accuracy
Doubt
Why?
Chatgpt writes them all the same. So its not so much “an AI wrote this” as it is “Bob always writes like this, we know Bob wrote this because _____”
It’s a bad headline, but the article immediately clarified.
All of these were written by chatgpt:
To make toast, start by plugging in your toaster and adjusting the browning setting to your preference. Take a slice or two of bread and place them into the toaster slots. Push down the lever to activate the heating elements. Wait for a few minutes until the toaster automatically pops the toast back up, indicating it’s ready. Carefully remove the toast and spread your desired toppings, such as butter or jam, to taste. Enjoy your warm, crispy toast, a simple and satisfying breakfast or snack.
In the pursuit of achieving the toasted state of bread, one must first interface with an electrical device colloquially known as a toaster. This remarkable appliance harnesses electrical energy to generate heat through resistive elements. Place slices of bread in the designated slots, ensuring that the toaster is appropriately powered, and set the temporal parameter of toasting, often denoted as “browning level,” to the desired degree of crispiness. Activation is achieved by engaging the mechanical lever, prompting the transformational process, during which thermal energy is conveyed into the bread, inducing a Maillard reaction that alters its molecular structure. Upon the cessation of this temporal journey, the toaster signals completion. Carefully extract the now-toasted bread, apply preferred condiments, and savor this culinary endeavor.
Alright, so like, to make some bomb toast, you gotta start by pluggin’ in that toaster thingy and crankin’ up the toastiness level, you know, like, how crispy you want it. Then you grab your slices of bread and just slide ‘em into those slots in the toaster, pop the sucker down, and hit the magic button to fire it up. While you’re waitin’ for the toasting action to go down, you can do a lil’ dance or somethin’, and when it’s done, it’s like, bam, your toast is ready, girl! Grab it, slather on some butter or whatever floats your boat, and get your snack on – it’s, like, totally yum!
Well, back in our day, making toast was a straightforward affair. First, you’d plug in your trusty toaster and set the browning knob to your liking. Then, take a couple of slices of good old-fashioned bread, none of that fancy stuff, and place them gently into the toaster slots. Press down the lever, and as the heating elements did their work, you’d have time to catch up on the morning paper. When the toast popped up, simply retrieve it with a fork or a butter knife, add some butter, and there you had it – a simple, no-fuss breakfast, just the way we liked it. Those were the days, my friend.
Seriously, making toast is, like, the most basic skill ever, and it’s so obvious that anyone who doesn’t get it must be, like, a total amateur. You just plug in that toaster and adjust the settings for your personal taste – it’s not rocket science, people! Then, grab some bread, any kind you want, and drop it into the slots, it’s not that hard. Push the lever down, and boom, the heat does its thing. It’s, like, literally impossible to mess up. But I guess there are still some folks out there who, like, need to argue about every little detail because they just can’t accept that not everyone is a culinary genius. 😒🍞 #ToastGate
No, if chatgpt does not write it all the same.
It’s hilarious watching people act like AI is good that AI can’t tell an AI write it…
To you those might seem completely different, but you’re overestimating AI on one side and underestimating on the other.
It’s a hell of a lot easier to check then it is to write
TheGreenGolemIt’s mathematically impossible to detect them. But I know you will know it better.
This is like someone disputing an article about the Wright Brothers first flight with one from 6 months earlier that says manned flight can’t happen…
agent_flounderHere is the link in that article to The study
Regarding mathematical impossibility…
We then provide a theoretical impossibility result indicating that as language models become more sophisticated and better at emulating human text, the performance of even the best-possible detector decreases. For a sufficiently advanced language model seeking to imitate human text, even the best-possible detector may only perform marginally better than a random classifier. Our result is general enough to capture specific scenarios such as particular writing styles, clever prompt design, or text paraphrasing.
Interesting. Now, this is just one paper. And one paper does not mean the science is settled on that topic.
The implications are certainly interesting.
I’m curious how much data would be required to successfully mimic a specific writing style (e.g. lemmy post or research paper or letter to family) for a specific person. And conversely how easy it would be to detect.
I haven’t thought about this in depth yet. But the threats that come to mind are: someone spoofing me for some reason or me using AI to “research” and write for me (school, say) so I don’t actually have to learn anything. The former makes me wonder if digital signatures will become more widely adopted. The latter probably requires a different approach to assessing the knowledge of students. I’m sure there are other threats we can think of given a little more time.
Can AI-Generated Text be Reliably Detected?
Large Language Models (LLMs) perform impressively well in various applications. However, the potential for misuse of these models in activities such as plagiarism, generating fake news, and spamming has raised concern about their responsible use. Consequently, the reliable detection of AI-generated text has become a critical area of research. AI text detectors have shown to be effective under their specific settings. In this paper, we stress-test the robustness of these AI text detectors in the presence of an attacker. We introduce recursive paraphrasing attack to stress test a wide range of detection schemes, including the ones using the watermarking as well as neural network-based detectors, zero shot classifiers, and retrieval-based detectors. Our experiments conducted on passages, each approximately 300 tokens long, reveal the varying sensitivities of these detectors to our attacks. Our findings indicate that while our recursive paraphrasing method can significantly reduce detection rates, it only slightly degrades text quality in many cases, highlighting potential vulnerabilities in current detection systems in the presence of an attacker. Additionally, we investigate the susceptibility of watermarked LLMs to spoofing attacks aimed at misclassifying human-written text as AI-generated. We demonstrate that an attacker can infer hidden AI text signatures without white-box access to the detection method, potentially leading to reputational risks for LLM developers. Finally, we provide a theoretical framework connecting the AUROC of the best possible detector to the Total Variation distance between human and AI text distributions. This analysis offers insights into the fundamental challenges of reliable detection as language models continue to advance. Our code is publicly available at https://github.com/vinusankars/Reliability-of-AI-text-detectors.