The "core reasoning" part of AI may be increasingly important to improve, and its "database of factual knowledge" aspects may be less and less important, maybe increasingly a hindrance. So more focused and specialized training may take over toward increasing reasoning precision, and not this never-ending stream of new data.
So maybe we'll get better reasoning and therefore better generated data/content in the wild, without this negative feedback loop everyone is worried about.
So maybe we'll get better reasoning and therefore better generated data/content in the wild, without this negative feedback loop everyone is worried about.