This claim was new to me recently, but I've seen it's become widespread. From other HN threads, it sounds like it's based on what Judge Alsup wrote in the Bartz v. Anthropic case, in which Anthropic's destroying scanned books was one factor supporting a finding of fair use.
I could see companies wanting to try to fit their activities into the same contours that the court approved there, since they could then argue that they should be just as protected as the defendant in that case was (that is, it may have inadvertently established a set of practices that companies would like to copy). However, it doesn't seem to me that the court concluded that destroying books was necessarily required (in particular because the contrary scenario wasn't examined there). That is, the court felt that destroying scanned books made the fair use argument stronger, but didn't say, as far as I know, that the fair use argument would necessarily fail without it.
But indeed, I guess it might take a lot for a company to decide to create a different and untested fact pattern and then defend it from scratch in new litigation.
> On the whole, people far too often hold strong opinions about topics for which they have limited information. This can cause social and political movements to have plenty of people behind them with limited ideas of what they are even fighting for.
If ads are integrated deeply into the responses, the model ceases to be a good model. It will adjust its output to best serve an ad, not best serve the user.
