The original source of the claim that ChatGPT overuses the word ‘delve’ seems to be AI Phrase Finder which makes the claim on the basis of “our dataset of 50,000 ChatGPT responses”. No more information is provided about this dataset. There’s also no information provided about what constitutes the ‘most common words’ within it. Presumably the most common words are ‘the’ (etc) which suggests they are using a different standard of what constitutes a common word. But they don’t say what this is!
This is not a reliable source which raises the question of why this claim spread as widely as it did. There’s an interesting bibliometrics debate which suggests this might not be 100% nonsense. But as far as I can see the claim spread through a network of (possibly AI generated!) content farms, as well as being credulously shared through social media. I’m not saying that ChatGPT doesn’t have tells, only that establishing them authoritatively is a methodologically complex undertaking which would only establish deeply imperfect results.
https://markcarrigan.net/2024/08/06/the-original-source-of-the-claim-that-chatgpt-overuses-the-word-delve/
#AI #ChatGPT #contentFarms #delve