@policytensor my guess is the Sunday/Monday "signing" didn't happen, was just a finalization of the text and announcement from Trump, first step to actual signing was alleviation of US blockade, then Iran was willing to officially sign
Iranian media didn't report official signing until now
@TylerAlterman that's the whole debate right now. clearly current techniques do great in easily verifiable domains (code, math). it is not clear if current techniques will adequately extrapolate to difficult to verify domains (have "taste", etc)
if only verifiable -> normal tech
else -> agi
@TheZvi feeling shooketh tbh. was holding out hope that current training techniques would not adequately transfer to difficult to verify domains - was not impressed with 4.7, 4.8, etc.
Fable is different. maybe this can go all the way...
@NunoSempere yea, except it gets kinda weird explaining female political preferences. this would imply it was always left wing but apparently that's more of a modern phenomenon so idk
@itaisher@AndyMasley this one is related to their claims re what this technology is potentially capable of and is evidence that they don't actually know what is possible here
not sure how seriously to take the petition, but it's very obvious to anyone reading in good faith that you are talking about standard academic practice rather than anything you do personally.
ppl can disagree about what should be allowed re citations and what penalties should be for mistakes/possible fraud, but coming after you is obvious witch hunt behavior and the worst of social media dynamics.
hopefully nothing comes of it and everyone can move on. good luck
fwiw i took that post to mostly be a threat/deterrence
i assume "recognition of [Iran's] sovereignty and rights" includes the Strait, and i don't see how the US officially agrees to that at this point (or ever, probably)
also note that post is in English - that account almost always posts in Persian, so a threat directed at American audiences tracks
@austinc3301 I'm not sure it'll be enough but my hunch is that their revenue will inflect meaningfully upward once they fully integrate ads
unless something dramatic changes I don't see how they can avoid ads
? Nash is about logical deduction, not Bayesian updating on evidence. Eg poker solvers play ~nash even though people empirically don't cuz it's too hard
also waitbutwhy guy posted "blue hanging on by a thread" when it was like 52-48, very possible it goes 49 blue by end and we all die if he doesn't ~coordinate. if we can coordinate then yeah blue seems good
I believe they are derivatives where the underlying oracle price is partly from most recent raise valuation/secondary price, and the other part is a time weighted average of the recent derivative price. The "mark" (current trading price) and the oracle are supposed to converge over time.
The idea is a liquid instrument that trades freely while having some tethering to "real" prices.
There is some (though not great) volume so I would take it as a low confidence best guess given current information. (also I may have misremembered or explained something wrong, the docs are available online to check details)
@gmiller@robertwiblin parents->children prove that it is possible in principle for a smarter, more powerful thing to care deeply about a dumber, less powerful thing
evolution did it through genes/kin selection. it seems possible in principle to instantiate something similar with machines
@Romy_Holland happy, cute faces = unlikely to eat or kill me = good, smiling, no crying. babies alert nearby allies to dangers to their survival by crying
@SarahTheHaider i found Nick Bostrom's book Superintelligence to be convincing for the general risk. also the audiobook has this british lady narrator which was perfect for the context and extremely listenable