-3

Are there any linguistic analysis algorithms designed to detect the output of undisclosed computerized large language models (bots) that pops up on question-answer sites, eg Quora? (vs text written by an individual human)

It would seem there is significant crossover between linguists and computer programmers. Surely some of them have developed basic principles of bot identification. Is any of it published?

4
  • 1
    You may want to ask this over at Artificial Intelligence (nothing about this question is about English). But the short answer to all of your questions is 'yes, but not in the way that you expect'. I feel like I've heard that some conference papers or arxiv have been published but only in the past year maybe? Which means people have considered it and have created toy systems. I'm pretty sure I've heard about attempts to -watermark- text produced by an LLM so it would be -easier- to detect LLM production.
    – Mitch
    Jan 4 at 16:33
  • 2
    Maybe this comment is of interest: english.meta.stackexchange.com/questions/15500/… : Pretty sure you must already be aware of it but here is a GPT2 Detector: huggingface.co/openai-detector Can't vouch for its effectiveness but is worth trying out. – Mohit Jan 4 at 5:16
    – Tinfoil Hat
    Jan 7 at 20:06
  • Looking more for linguistic models
    – fertilizerspike
    Jan 7 at 20:18
  • 2
    at_fertilizerspike, How is what you're looking for, what you call a 'linguistic model', different from what is called by the community a 'language model'? I say that because @TinfoilHat's comment sounds exactly like a (correct) answer for your question, a deep learning program that on input of text, outputs a yes/no was the text generated by something like GPT. That's a different situation from using some general features of the text (that a person figured out and listed).
    – Mitch
    Jan 8 at 20:03

1 Answer 1

Reset to default
0

It is early in the days of useful services (websites that take in text and output probability that it was produced by an LLM) but things are changing fast. Here is a list of some usable sites and some academic papers made be Sebastian Raschka:

ChatGPT is the latest craze (Nov 15, 2022) and like many AI successes (qualified), it is not the first to do what it does, it's just packaged well. So likewise for many of these detectors - while a lot of things have happened in the past couple months, people have been working for years on getting these detectors available seemingly at the same time as ChaGPT came out.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .