And I should have spoken up to defend you then – sorry. But the people chastising you weren’t the ones who draw the lines for COG. They’re a chunk of the audience with strongly-held views, and you should gauge their feedback accordingly.
I understand how interacting with the strongest views on this might lead one to
but I’m certainly not going to do that, and I don’t think anyone else should be forced to.
I agree with Harris that the strongest reason to avoid ChatGPT for research is its unreliability. I’ve yet to see AI research output in areas I know anything about that didn’t include significant inaccuracies, even if not outright hallucinations.
ChatGPT being trained on the moral equivalent of theft is another reason to refrain… but as I noted above, while I think that’s more than sufficient reason not to pay for it, I suspect that non-subscriber use at this point at this point hurts OpenAI more than it helps them. So the best action to take on that basis is debatable.
There is plenty of non-copyrighted, open-source content out there for purposes of AI training, and my understanding is that it was adequate for loads of machine learning functions, including Google Translate. LLMs have ended up demanding more writing than has ever been produced in all of human history, and a bunch of companies have responded to that by pushing “fair use” past what ought to be a legal breaking point, if not ignoring it entirely. But not all AI is based on LLM-style training.
CoG’s caution about Google Translate is based on the inaccuracy of its translations; the rule isn’t “don’t use it,” it’s “don’t rely on its output without checking with someone fluent in English.” They note that it works OK for short articles, which is not what you say if you’re ruling it out entirely on moral/legal grounds.
Their caution about Grammarly isn’t about the spell/grammar check, which they call “fine,” but about using its specifically generative functions. That can lead to the AI (re)writing large chunks of text for you, with the ensuing problems around “sounding like an AI” and not being copyrightable which underlie the CoG/HG policy of not publishing text generated by AI.
With Harris, I think that line’s pretty clear. There’s a vocal share of the CoG/HG audience who would like the line to be drawn at “you must never have used any AI for anything, ever,” but that’s their line, not CoG’s, and should be treated like the opinion of any other chunk of the reader base with a strong opinion.