@Ted_Underwood Do you think that's contingent on on a continuing advance of SOTA across the field? B/c what i've found weird about the last three years is how different advances have looked than what would have been foreseen.

@Ted_Underwood Like, I wrote a paper in 2016 about dumb methods for embedding books on the *assumption* that smart methods for long-document embedding would exist in the next few years--but that still doesn't exist! Conversely, I'm shocked by how much text generation has advanced...

@Ted_Underwood But it's not really clear to me how much text generation has to do with anything I have a professional interest in. Ditto image generation...

@Ted_Underwood So I just keep coming back around to the idea that the most important contribution in this area involves the creation of new labeled datasets from messier sources and more effective discovery engines--but humanities/social sciences tend not to cite either.