I glimpsed though the article and I didn’t see it.
It said it used frequency of words for determination of AI usage.
Did it consider that AI is mostly trained of those papers and thus AI detection methods could not be as accurate as thought by the ones making this analysis?
Did the just considered general “generic AI repeated words” or did the actually used a frequency analysis of the actual fingerprint left by the LMM that companies put there for forensic analysis (and that I don’t ever know of the companies are publishing the signatures they use).
I glimpsed though the article and I didn’t see it.
It said it used frequency of words for determination of AI usage.
Did it consider that AI is mostly trained of those papers and thus AI detection methods could not be as accurate as thought by the ones making this analysis?
Did the just considered general “generic AI repeated words” or did the actually used a frequency analysis of the actual fingerprint left by the LMM that companies put there for forensic analysis (and that I don’t ever know of the companies are publishing the signatures they use).
LLMs are not, in fact, trained on “mostly those papers”