@meowcat_x41@RichardHanania There's plenty of value to knowing more in the many thousands of days (likely) between now and widely usable/affordable/safe elective BMIs.
@AussieSamizdat@RichardHanania Sounds like your intuitions are accurate to free-tier AI from last year, perhaps not as much frontier-tier today. It still has flaws and makes errors ofc, but far from constantly, and honestly it's better than most decent and higher quality human sources by now imo
@sneaker_eagle@RichardHanania You can test your LLM based learning where it's relatively checkable, then use same methods where it's less, kinda like AI training lol
@panickssery@davideoks@robertwiblin People care about their future selves in old age or laid off, their family members, etc. Those same people are mostly contributing economically in essential ways, so elites can't simply crush or ignore them.
Admittedly, food stamps are harder than entitlements under that model
@JoonGTP@geoffreyirving Stopping yes, slowing or pausing? No, it's not actually naive at all. We do it to all kinds of technologies on a regular basis, and AI relies on massive data centers you can see from space filled with expensive compute from a hyper-specialized global supply chain
@roystgnr@4rcherhume They didn't say "fast" takeoff, but also, obviously things are speeding up. And while slow takeoff did originally mean ~what's been happening, that's not really how people use it anymore in my experience (not saying that's good or bad)
@4rcherhume@bicanis_baddog 6 months should be enough, maybe 3 or 4 lol. Though I disagree with the "Climate Change denier" framing, we're not *quite* there yet.
@the_yanco@FBI I mean Jihad seems almost impossible, not sure something like AI 2040 is less realistic and it's certainly better. Or like passing the Bernie Superintelligence bill (which won't happen), would be better + easier imo.
Viva la revolution is virtually always wrong (in liberalism)
@StiberAdam@ElderTyrannus@jeremiecharris This is definitely very interesting, and I should try it more, but I don't think it changes the basics of communication. If a persona/tool obfuscates your verbal transparency, it's probably either not right for a particular context, or necessitates a translation step, etc.
@StiberAdam@ElderTyrannus@jeremiecharris I mean maybe in very specific contexts, but mostly I'd say respecting the reader includes trying to bridge from your social and intellectual context to theirs. And just generally treating words as an effort to clearly communicate; the alternative is Jordan Peterson
@2025Blog@tracewoodgrains Yeah, I think he's wrong but not for that reason. It's the expectation that a transformational tech will be widely + strongly transformational in only a few years. Things have been proceding far faster than with early electricity generation or the internet in basically every way
@Nick65979825@SashaGusevPosts@tracewoodgrains E.g. In terms of pre-industrial revolution rates of economic growth, the transition was quite fast: "only" 1-2 centuries. AI will increasingly speed itself up in terms of research, diffusion, automation, etc., so should also be fast relative to current day rates: only a few years
@Nick65979825@SashaGusevPosts@tracewoodgrains 5-10% GDP growth of frontier economies would absolutely be huge/not disappointing, but only if it happened quickly enough (e.g. early 2030s, with clear trend break ramping up by the end of the decade). And that's what we will get if it's remotely as fast relative to pre-industry
@NelsonEdmundb@MeekFrances@tracewoodgrains Wouldn't the first LLM be closer to ARPANET if we're comparing modern AI to the internet? So 2018-2026, or 8 years, thus 1982 internet? It probably doesn't make much of a difference, but you might as well be accurate if you're relying on precise numbers like that
@Viresh__R@Marcus_J_W It's just efficiency from RLVR, compressing more info into fewer tokens. A bit concerning but it ultimately does little to reduce comprehensibility or monitorability unless I'm missing something