faceh
2d ago
Tend to agree.
Also, part of the issue is that a job can very much look like bullshit right up until some extremely important necessity arises.
Some amount of 'busywork' is there so that someone can stay occupied while they're being paid to be present in case that [event] occurs, which can be at almost any time, and the work has to be easy and unimportant enough that they can set it aside to attend to the event without something else catching on fire.
Rough example, the security guard at the bank might sit around watching videos on his phone for most of a year, but he is expected to jump to it if a guy with a ski mask appears.
So glad my alter ego already posted this, saved me a lot of hassle writing my own response.
Especially this:
This commoditization vector is where the actual bear case lives. Forget your framing about demand evaporating with the busywork. The version of the worry I'd take seriously has total inference going up 100x while AI-provider gross margins compress to nothing because the underlying capability turns out to be fungible across providers.
Right now, this is where I predict the LLMs will end up if the exponential growth curve does taper off and become sigmoid before we hit AGI. Intelligence will become akin to a utility. Literally, tokens will be treated in the manner of drinking water or electricity or internet data itself. It'll just be expected that every individual and business will have a hookup and they'll pay a monthly bill for their usage, the price of which won't vary much between providers, and where the ease of switching providers is practically instantaneous.
Doubtful it'll become a public commodity though.
The somewhat close analogue is Bitcoin Mining. Remember it used to be viable to mine on CPU, then GPUs were the only method, then ASICs. And now, as far as I can tell, mining power literally just sorts out to where the cost of electricity is cheaper/subsidized, and its pointless to try to compete if your power costs even 5% more.
Although I have to imagine, similar to electricity prices, there'll be some dynamism in it, with prices potentially shifting not just due to the cost of various inputs, but the shifts in demand in various geographical areas.
Hah, I wonder if there'll be the bargain-tier option to set your agents to only run when there are lapses in demand.
If this does happen, it should strongly inspire a tech race into cheaper electricity generation. A method for converting electricity directly into usable intellectual work is the sign of the next industrial revolution. That's exciting.
You use Mythos or Opus for the demanding work, and smaller models where quality doesn't come first.
This is my other thought. We're going to get a severe tier system for model 'intelligence' and some protocol for determining which model to use for given tasks based on complexity/importance. The top tiers might be the equivalent of Deep Thought from Hitchhiker's Guide where it takes them immense amounts of time, at serious expense, to compute their answers, but said answers are guaranteed to be correct regardless of the complexity of the question (but make sure you specify the question enough to understand the answer). The bottom tiers might be able to assist you at Bar Trivia when you're too drunk to remember movie titles.
So yeah if things taper off before AGI, I expect we'll get some intelligence that is too cheap to meter, but the good stuff will only be available at Top-Shelf pricing.
My overall take? The big guys want to be first to AGI, then hope that RSI takes them all the way to ASI and incredible wealth.
But this is the driving force behind the big bets, all evidence is that the big players believe the hype is real, and the prize for winning (or, at least not losing) is so immense that they don't know how to rationally calculate for it.
faceh
2d ago
Basically this amounts to an expectation that the current 'frontrunners' are vulnerable and some as-yet-uncertain player will succeed in disrupting their runs.
Which I can see as a possibility, but I'm not sure where to look for them.
faceh
3d ago
I think I take your meaning, but allow me to try and parse this:
The markets are expecting a Obama/Trump style moment where someone unexpectedly comes in
The markets expect that the unexpected will happen. I guess its just a wait and see thing, account for the uncertainty.
And as much as Obama provides good precedent, I think the subsequent years where Hillary and then Biden locked down the nom against upstarts (including Bernie... twice) makes it less likely.
Its frankly hilarious to me that the Dems are basically stuck with the coalition they built and all its dysfunction because they elevated the AOCs, Kamala Harris', Jasmine Crocketts and Stacey Abrams amongst them, and the most motivated and active parts of their base are all-in on identity politics, so trying to wrest back the controls will require exercises of raw, naked power that is just as likely to bite them in the ass as it is to select a viable candidate.
My biggest fear for the GOP is post-Trump (or the small chance Trump simply declines to leave the stage) power struggles and the constant tendency at infighting at the most inopportune times.
But the GOP now has a deep qualified leadership bench by comparison, and one that isn't (as far as I can tell) beholden to constituencies that will demand the faucet of grift, graft and fraud be kept all the way open or they'll revolt. Which is to say, a GOP official can actually try things that might make things better.
it is quite clear that as of now the USA is in no position to contest the globe with the likes of China or Russia.
...
...I don't think it is 'quite clear' and the case laid out here just takes it for granted.
I'm sorry was there not a whole thing where Russian and Chinese anti-air completely failed to pose a real threat in Venezuela and later in Iran?
Did I hallucinate a whole news story about an incursion into Venezuela without even needing to perform a significant amount of bombing? Last I checked Maduro is still in prison in New York.
Was that not a notable development that suggests that U.S. air superiority is even more lopsided than might have been assumed?
I'm actually asking for a chain of reasoning to explain how you came out of the last 5 months with a higher estimate of Russian and Chinese military efficacy compared to the U.S. than previously. This viewpoint befuddles me.
I don't know what your precise definition of "contest the globe" is but holy cow, if nukes were off the table, how exactly do those countries even show up for the competition, let alone win it convincingly. What exactly is the edge you're giving to them that will outperform demonstrated U.S. capabilities, before discussing the stuff that is as-yet undemonstrated.
Find it annoying to see these posts that basically affirming the consequent by bolstering a given conclusion they've reached (the U.S. failed to completely demolish Iran's nuclear program) then pretending their premise (U.S. military capacity is hamstrung under Donald Trump) without a supporting scaffold of logic. We're barely two years into Trump two, after four years of whatever the hell Biden was, and making bold prognostications about his efforts being failures kind of belies the results we've seen overall.
which’ll hopefully initiate a process of rebuilding for the damages the last decade of Trumpist politics have inflicted on the nation.
Can't help but notice some motivated reasoning there, bub. Especially because there's that glaring interruption in "the last decade" of Trumpist politics that goes utterly glossed over.
Curious to mention Kamala in your post but I do a ctrl+f "Biden" and its empty. Howboutdat.
Anyway, sorry to sound flippant, but it really helps that when you're trying to advance a particular conclusion, you actually back up the premises you're using that are the most controversial or extraordinary, such as "a decade of damages" from Trumpist politics (damage inflicted where? of what magnitude?), or "...that his rule has largely been to the detriment of... America’s relationship with its allies in NATO" (As compared to before? What precisely has deteriorated?).
Anyhow:
Rubio or Vance remain the favorites for winning the 2028 Presidential election Much can change in between, but the assumption that there isn't a viable path forward that actually preserves much of the Trump coalition seems wishful, there's likely a power struggle to come, but all those Trump-backed challengers just blew out a bunch of incumbents in Indiana, so it seems pretty clear where the center of gravity within the GOP lies right now.
Oh, also, on the manufacturing front. Upward momentum after decades of decline... suggests something has changed or improved. Yes, that includes the Rust Belt.
I dunno, this analysis seems to be a little undercooked.
Tend to agree that 4 was the one where everything they'd learned came together and had the fewest annoyances overall.
And it achieved strong variety of gameplay and challenges, and the open world and fetch quests didn't feel too tedious 90% of the time.
Indeed, can't think of any part of it I genuinely dislike, other than the radio host guy, and mostly because of how repetitive he was.
I'm still desperate for a Far Cry 2 remake that adds more combat variety, and some of the quality of life upgrades from the subsequent entries, but keeps the core, raw immersive difficulty elements.
FC2 remains one of the few games I've ever played that balances out the "One Man Army" power fantasy with the "actually you can very easily die because you are all alone, in a hostile environment, with crappy weapons/tools and aggressively outnumbered by people who want you dead." Which is to say, it natively encouraged 'sensible' strategic and tactical approaches and punished you if you got too wild. At least, outside of explicit survival/horror games.
Also remains the best AI enemies ever had in the series... and in almost any other game I've ever played.
I have a distinct memory of aggroing a small group of enemies at an encampment by a river. I shelter in a small shack, firing out the windows. The enemies actually started flanking my position and then, to my absolute shock, lobbed a grenade right through the window practically into my lap. That was a tactic I would attribute to a human player. And I still rarely see that done by AI in even more modern games.
They were trying to flush me out of cover so they could mow me down.
Also the only game I know of where injuring an enemy and waiting for a comrade to come and rescue him so you can kill both was a viable tactic!
I've read my Lesswrong and I find the Yudkowskian arguments convincing enough to believe we're going to eventually hit the "foom" point even if progress stagnates in the short term (which it hasn't, as you note).
An AI with Von Neumann level intellect that is able to self-replicate and cooperate with its copies AND has access to its own source code should, I'd think, be able to solve most bottlenecks to its ascension in the course of a day.
I do not feel remotely qualified to guess what the actual tipping point will be.
More options
Context Copy link