3 ms·
That is something I have been wondering about LLMs. Do they keep track of large chunks of text, they were trained on? e.g., Does ChatGPT know that "Four score
by rawgabbit 3y ago
That is something I have been wondering about LLMs. Do they keep track of large chunks of text, they were trained on? e.g., Does ChatGPT know that "Four scores and seven years" was a speech given by Abraham Lincoln to commemorate the battle of Gettysburg? Or does some articles say ChatGPT just write "plausible" sentences.
In other words, the reverse ChatGPT that tries to identify papers written by AI. Is it really sure that the text is plagiarized or is it simply "plausibly" written by AI. That is if something like this goes go to court and the judge asks serious questions... will these AI detectors be laughed at?
- thfuran 3y agoBased on the long history of taking complete nonsense very seriously in forensics, I'd say the odds are pretty good that it'll be taken way too seriously, at least for other AI systems, but perhaps not specifically plagiarism detectors if a large chunk of the jury has personal experience fighting with overzealous anticheat crapware at school.