@ug0tpwn3d@todaymare it worked fine (on linux!) when I had it, except whenever there was an upgrade it randomly went into some sort of disaster mode or other. and also you had to use a magic two year old version of a package to get full performance. but when it worked it was pretty fast.
@UnderwaterBepis in my framework I have a tool call that lets it compact results. basically a whole snippet of context is replaced with a summary that it writes itself. seems to work well. some toolcalls with lots of output are also ephemeral, meaning they're only included for the next request.
@MasterTimBlais it can't be substrate dependent because there's no law of nature that mentions consciousness.
*anything* computational is substrate independent.
@RandomSprint to be fair, they shouldn't enter them like that. a lot of american election conspiracying is the result of bad data entry practice with very very predictable consequences. it's important for the count to be right, but then it's just as important for that rightness to be legible.
@Kukia_Ch@sanson_ganbaru wikipedia, shortened: "Kohlrabi has been created by selection for a swollen, nearly spherical shape; its origin is the same as that of cabbage, broccoli, cauliflower, kale, collard greens, and Brussels sprouts: they are all bred from the wild cabbage plant."
Jesus, that's a lot.
@real_faox@Jbn3ex that said if your stance really is "rape is 100x worse than murder" then I'm not sure we're disagreeing at all, except in that I'd still argue that it should be emphasized that both assaults are highly unlikely outcomes and that meeting men while hiking is generally safe.
@real_faox@Jbn3ex I mean that's a fair point of view, it makes sense as a zoologist that you have better odds on the bear encounter. I just think that you underestimate the degree to which male sexual assault is also situationally predictable. (ie. overwhelmingly not in this situation.) 1/2
@real_faox@Jbn3ex I mean I assumed they were both drawn from the ordinary distributions for woods encounters. that *is* the context for a woods encounter. I feel like you give the bear a lot of credit and mostly ignore the fact that men also have massive stratification in terms of risk.
@real_faox@Jbn3ex Yes, what I'm saying is if the bear and man are drawn from the usual encounter populations for hiking in the woods the bear is still about two orders of magnitude more dangerous to you, so you have to *really* hate sexual assault (100x more than murder!) to make it even.
@real_faox@Jbn3ex Sure but the vast majority of those situations are domestic or urban. I'm not disagreeing in general, I just think you underestimate how much the framing of the original experiment downweights the risk factors.
@freed_dfilan I use LLMs very heavily and am a doomer. My view of the current state is, it's great at building new software but it's not great at building great software.
My theory is great software requires taste, and you don't get that without *very* long-term (>context) reward attribution.
@real_faox@Jbn3ex to be clear what I'm saying, *worrying in advance* about strange men in the woods more than bears makes sense (though not worrying more than about partners/friends). But if you're *crossing the path of a bear*, do be considerably more worried than if you're walking past a man.
@real_faox@Jbn3ex The sexual assault statistics are massively skewed by situation. If you meet a man in the woods you are actually in one of the *safest* situations in terms of risk, because hikers skew affluent/educated and violent crime is mostly a low-intelligence phenomenon.