Super interesting stuff so far! It seems that quite a few of the worries (particularly in “Unclear definitions and constrained research thinking” and “clarity”) seem to stem from AI safety currently being a pre-paradigmatic field. This might suggest that it would be particularly impactful to explore more than exploiting (though this depends on just how aggressive ones timelines are). It might also suggest that having a more positive “let’s try out this funky idea and see where it leads” culture could be worth pursuing (to a higher degree than is being done currently). All and all, very nice to see pain points fleshed out in this way!
(Disclaimer: I do work for Apart Research with Esben, so please adjust for that excitement in your own assessment :))
Super interesting stuff so far! It seems that quite a few of the worries (particularly in “Unclear definitions and constrained research thinking” and “clarity”) seem to stem from AI safety currently being a pre-paradigmatic field. This might suggest that it would be particularly impactful to explore more than exploiting (though this depends on just how aggressive ones timelines are). It might also suggest that having a more positive “let’s try out this funky idea and see where it leads” culture could be worth pursuing (to a higher degree than is being done currently). All and all, very nice to see pain points fleshed out in this way!
(Disclaimer: I do work for Apart Research with Esben, so please adjust for that excitement in your own assessment :))