Open-ended data is a bottleneck for AI systems tackling real knowledge work

I think the issue with open endedness has been getting reliable training signal in the first place. Not sure how easy it is to systematically capture that signal, but cool to see teams are working on this

This is an ad.