You could argue there's now more niche (coding) knowledge than ever before. Do we seriously think that "de-identified" enterprise data isn't making into the training sets of LLMs? None of that proprietary code would had ever been posted to SO, but it's being posted millions of times a day to Claude and ChatGPT.
You could argue there's now more niche (coding) knowledge than ever before. Do we seriously think that "de-identified" enterprise data isn't making into the training sets of LLMs? None of that proprietary code would had ever been posted to SO, but it's being posted millions of times a day to Claude and ChatGPT.