@TradMaldOwl@AlexGodofsky@BobMurphyEcon@markeburge I disagree - fairly natural heuristics like 'try a smaller version' lead pretty directly to the answer. (I solved this ~immediately when I first encountered it. I've asked a few friends for fun and they mostly solved it pretty easily too)
@johnvmcdonnell@xav_moss Xavier, John was critiquing you for seemingly objecting to discussion of tradeoffs, not weighing in on the object level debate (which you've also misrepresented - lyman was critiquing EA for prioritizing saving lives over longer term development; neither side was pro-killing)
@xav_moss@steady_drumbeat@colintj@KelseyTuoc Are you saying Kelsey shouldn't have argued against Lyman's position on GDP, and/or that she should have done so in some other way (eg more 'concretely')? If the latter, can you be more specific?
(I disagree w you but am asking out of curiosity, not to argue)
@TheStalwart@LDrogen And even this overstates it, imo - the public reasoning (iirc) was (1) this is probably fine, but it's new, let's be cautious, (2) we are not actually so worried about GPT-2 but about more powerful future models and want to practice and build prudent institutional habits
@eschwitz Aaah, missed this by a few hours! Have been looking forward to reading this since I heard you were working on it... Any way to get a copy faster than waiting for the hard copy? (Don't mind paying)
@RichardHanania You seem to be missing Scott's point here. He agrees there was too much worrying, and thinks this is partly downstream of optimists losing credibility due to their initial dismissiveness.
@adam_kranz@StatedClearly Fair enough, but this reasonable version doesn't seem to support the initial claim. The language of physics doesn't have (concise) vocab for systems at higher levels of description. This seems independent of causal closure. (And also, to me, like a merely semantic claim.)
@emlwaters@LeahLibresco@madeleinekearns FYI it's much shorter than you might be expecting for an ACX post! (In case you hadn't read because budgeting for a longer block)
@0xen1@KirkegaardEmil Haha, yep - though to be fair it did accurately capture that I wasn't familiar with 'piquet' (and also my questionable meta-judgment that 'millinery' may not have been close enough). Seems fair enough!
@JeremiahDJohns I agree with the others - Anthropic currently has the lead, so if doing just one, that is probably the way to go. But $20/month to each is totally worth it, if only to figure out your own preferences