The current architectures fundamentally aren’t capable of such complexity, no matter how big they get or what prompt wrappers they have.
There are some interesting, deeper innovations in papers, but the AI hyperscalers seem to have little awareness of them, and I’ve seen so many cool experiments just drift by with no further testing these past few years. Which, again, suggests whatever approach the purse holders are taking is not a “innovate our way to better complexity” one.
They certainly do if they have to draw cash at a loss after they, say, lose a job from a crash. Or draw because of inflation like we’re at a huge risk of now.
Other times, it’s big funds (like retirement funds) folks are “automatically” invested in that make some bad forced decisions during a crash.
You’re right, there is an unfortunate tendency to panic withdraw during a crash (that’s the idea, right?), but it’s not always a choice.
It also depends what you mean by “hurt.”
Nvidia/AMD stocks, for example, are going to drop like meteors, but the physical companies themselves will be fine. They’re like the picks and shovels makers of the California Gold Rush; they’ve made their piles of cash and will go back to business as usual. Hence, I’m not selling my long-held AMD, even though I’m certainly not buying more. Yet.
And some totally unrelated companies may be disproportionately hurt by the pullback of a serious recession.
One comment I will make on LLMs specifically is it’s more of a “race to the bottom” than you’d think. Between sparsity research (with the recent MoE trend being a rather crude stopgap if you ask me), alternate attention schemes, finetuning advancements, computing shifts like BitNet all in the pipe, and all the open models from China and others, well…
The end point feels like local inference of specialized, freely licensed models. As useful, niche tools, not superintelligence.
They’re low power basically free, hence seemingly inevitable.
That’s utterly apocalyptic if you’re someone like Sam Altman or Jeff Bezos. OpenAI can’t make money off that, which is why they’re lobbying to kill it and preaching infinite scaling that won’t work.
Normally I would agree.
But the weight of this one obviously hyped sector is measurably, historically huge: https://www.apolloacademy.com/wp-content/uploads/2025/09/ExtremeAIConcentration-090825.pdf
With a lot of “circular investment” reminiscent of previous bad behavior: https://www.axios.com/2025/09/25/nvidia-openai-investment-ai
Obviously don’t sell after a crash, or sell the absolute least you can to live; that is rule #1.
…But I think it’s prudent to save a bit extra and shuffle some investment out of the S&P 500 pre-emptively, as it’s starting to resemble an AI evangelism hype fund. I’m not that old/experienced, but I’ve never seen anything like this in the market, especially from my perspective in the ML tinkerer community where, ironically, it’s obvious how much this all stinks. All the academics know it.
Saving to invest is hard, sometimes. And finance knowledge/attention is finite. Some folks are at risk of drawing from investments after a crash.
And not everyone here was invested during the 2008 crash, and may have only experienced the ultra-bizzare COVID rebound.
I don’t mean to be rude, but there are a lot of legit reasons most of Lemmy is probably not into ultra long buy-and-hold investing.
Real estate is a trouble prone investment normally, much less in this crazy market; I specifically wouldn’t want to touch that right now.
Can’t speak for metals, but also be careful there…
Thing about a bubble like this is you don’t know when it’s going to pop. I like the saying “the market can stay irrational longer than you can stay solvent.”
What I’m saying is to be careful about going all in on more pure hedges. If this lasts another 4 years and one’s into stuff like XMAG and metals, and they drop in a crash anyway, you may end up in a worse position than if you had held the S&P 500. I think a better perspective is to avoid “buying a hedge” and instead invest in companies (or other assets) one thinks will be productive and grow with the bubble or not. They’ll grow however long the bubble goes, and keep growing after.
And I am convinced that LLMs are fundamentally incapable of delivering on the promises being made by the AI CEOs.
As a, uh, atypical American, and someone into the ML scene and previously employed in an LLM dev job… I agree.
I don’t think ML is going away, as what’s been made so far are niche tools in the same way a hammer is, but the level of hype and conning is literally criminal.
If you can shift stocks around, take them out of indexes and put the cash in crash-resilient stocks like Berkshire Hathaway (which somewhat famously/infamously saves cash to buy dips during crashes), or Walmart. I’m thinking on such a “Noah’s Ark” basket for myself.
I’m not knowledgeable enough to comment on bonds, gold, or whatever else your savings may be in. But don’t believe a word anyone says to you about crypto.
Start saving a bit extra too, if possible, as the crash may not come for some time. And you want to avoid selling invested savings when the markets at its lowest.
On the tech side, you can get more into self hosting to not be so dependent on Big Tech. You’re on the perfect site to learn that.
If you ask me, that even includes dabbling in open-weights ML stuff, as that might suddenly become a more marketable skill once all the OpenAI hype implodes, and companies sipping the Koolaid turn more practical/frugal.
Other than that… I dunno. Depends on your work and lifestyle, I suppose. I think this will be a bumpy ride no matter what we do.
And that’s why Sora sucks, because it’s censored, closed weights, totally opaque, largely toolless, and peddled like spam.
I mean, I’m as big a ML fans as you’ll find on Lemmy, but this is a slop machine to build some Altman hype.
A controllable, integrated version as a tool, with augmentations like VACE or SDXLs controlnet would be neat. Thats also great because it’s not so easy for 1 click zero effort automated spam, which is by far Sora’s largest market as is.
…And guess what. We have that, it’s neat already, it’s open weights, it’s improving, and it’s not so controversial/abused because there’s an actual tiny barrier of entry to using it, like Davinci Resolve vs instagram filters.
Yeah, I’m pretty sure working Nvidia on wayland is a very recent thing.
Honestly I just boot from my (AMD) IGP for linux, which is better for compute anyway.
I guess it depends on what that ‘something’ is.
My linux partition used to be like this (mostly Nvidia issues), but its been relatively well behaved. And now my Windows install has become a pain with UWP apps, printers, and LAN drives, specifically, that I’ve just given up trying to resolve TBH. Not to speak of some programming stuff.
Both OSes are tools that make specific things easier.
So many folks seem to be the opposite of me…
Linux just works now. Shit with my printer, device drivers, LAN things, stuff like like is like wrestling an animal on Windows for some reason, and… just works with KDE. It’s like they’ve swapped places.
Random Windows apps works better in wine than they do in actual windows, sometimes. With no fuss: I double click and they launch, that’s it.
Don’t even get me started on security.
But Linux is (mostly) not performant for gaming, at least not on Nvidia. It’s… fine, but I’m not going to take a 10%+ hit, sometimes much more severe, and poorer support for HDR, frame limiters, mod tools and such when I can just boot neutered Windows instead.
So I’m not getting away from Windows in the near future, but to frank, I don’t understand why more folks (who get past the admittedly tall hurdle of learning about partitioning and installing an OS) don’t dual boot, or seek to use certain poorly supported Linux native apps when double clicking exes mostly just works.
But my point is you don’t have to pick and choose. And there’s no commitment. You can have your cake and eat it, and send the cake back if you don’t like it.
No, I just skimmed the transcript because it’s an hour long, heh.
I did get that bit about SETI and the original paper, which is interesting, and also agree that astronomers looking for them over the paper is hilarious and stupid.
Man, I miss my jailbroken iPhone 5.
It was like having your cake and eating it, and somehow its stock (much less tweaked) UI is less clunky than whatever TF Apple has done to my discount 16. Maybe it’s because I was using Android in between, but still…
With the caveat that I only read the transcripts, I don’t find that compelling at all.
The initial sentiment is correct; folks like Sam Altman responding to existential problems like “oh we can just build a Dyson Sphere in 30 years” should be in freaking jail instead of power.
But the only other justification I see is “well, this is stupidly impractical in the context of current humans.” Things like:
“What, we make all those nanobots and get all that energy with fusion and use it to disassemble Jupiter?”
“Why don’t we just use that energy to leave the solar system?”
“Say it’s a Dyson Swarm; what do we do living on all those solar satellites?”
She’s fallen into the same trap of “existing sci fi” she accuses other of falling into.
We’re not talking about a bunch of people in space looking to expand a habit. At this point, we’re talking about some AI that’s already converted an entire moons worth of mass into computronium, can upload folks to VR and simulate realities, that can reconfigure atomic nuclei into ultradense strings of matter or construct and control tiny black holes to generate energy and elements.
It’s left the solar system loooong ago.
Its capabilities, needs, and goals are completely umhuman, and at that point pondering how to efficiently capture the output of all this stellar mass sustainably is absolutely practical to plan. A Dyson Sphere (or more practically a swarm) isn’t the only way, but it’s not the worst idea for a “young” intelligence. And in OA, at a certain point, the Sephrotics seem to construct “sci fi” dyson spheres as habitats for aesthetic reasons, whereas their actual industrial/computational bases are more utilitarian arrangements of masses.
+1, I get it.
I don’t know how to say this inoffensively, but I think China (speaking broadly) has a cultural victim complex, which is understandable given their history.
And I think China and Russia governments stoke this victimization for political benefit, kinda like the US convervative movement is doing now.
Anyway, I think that leads to reflexive denial of their own atrocities as if its an abuser blaming a more enlightened victim. And as for the tankies actually outside of those countries, well… I don’t really know.
So what have tankies got to say here?
They seem real quiet.
Yeah, I buy the filter (or at least a big filter) being early. That does seem like a freak accident, even with all that time for it.
But on the spread of civilization, this is why I love Orion’s arm: it posits that if a civilization like ours makes it another few thousand years, it’ll expanded in a bubble at a significant fraction of the speed of light and be extremely difficult to extinguish at that point, meaning civilization should have spread across galaxies by now:
https://www.orionsarm.com/eg-article/49333a6b7d29f
That makes a lot of sense to me.
And the fiction, even as wild as it is, gives the still somewhat unsolved Fermi Paradox a lot of thought:
https://www.orionsarm.com/eg-article/464d087672fe7
I particularly like the ‘Ginnungagap Theory’ that, perhaps, there’s some unknown barrier to expansion.
https://www.orionsarm.com/eg-article/464e942db2789
“Even with all the equipment available in the Civilized Galaxy and beyond the amount of the Universe which can be examined in detail is tiny. Imagine our own Galaxy as a deep sea fish, with very sharp but tiny eyes, peering at the other galaxies with trepidation.”
I have, but I’m also concerned that humanity got “lucky” so far and that this won’t happen again. There are theories positing that there are several blocking “gates” to civilization, and humanity passed an exceptional number of them already.
It’s reasonable to assert that’s a misleading, human centric perspective; but I’d also point out that the Fermi Paradox supports it. Either:
The conditions that gave birth to our civilization are not exceptional, and spread intelligent life is hiding from us (unlikely at this point, I think)
They are exceptional, and we just happened to have passed many unlikely hurdles so far (hence it is critical we don’t trip up at the end here).
They are not that exceptional (eg more intelligent vertebrates will rise, and would rise on other planets), but there is some gate we are not aware of yet (which I have heard called the Great Filter).
Another suspicious coincidence I’d point out is that we are, seemingly, the only advanced civilization from Earth so far. If we died out soon, other vertebrates that rise would find evidence of us by this point, wouldn’t they? Hence odds are we wouldnt be the first and we would have found precursors if ‘vertebrates rising and then killing themselves off’ was a likely scenario.
TL;DR: I suspect vertebrates -> our tech level is a difficult jump.
Well Musk’s quote there isn’t wrong.
Musk isn’t an engineer, but neither are those consultants. They should all be deferring to deeper within the company for technical decisions instead of conflating some sci fi literacy with practical decision making, and leave musk to making much coarser ones at most.