You probably won't be able to run a sota model on a single Talaas card. You probably need a set. It should theoretically be cheaper, faster, more power efficient than Nvidia at running that model. But then you are stuck with a chip that can run only that model. And I understand that context size will also be a problem.
Yeah, that was why I felt that, while Taalas might not be the right thing in general, it could be really appealing in a laptop, that would otherwise not be able to run any big model at all.
I think I misunderstood what the technology could do a little bit, because I imagined these fixed matrix multiplies encoded into hardware as small-ish, as using routing space etc., not just fast.
I think what I really imagined was useful would be more easily achieved by having a big ROM with model weights connected to an ordinary GPU.
I like this observation very much. Intellectual deflationary spiral is exactly how it feels at the moment and we might end up in a worse place because of it.
This line of thinking has baked in the assumption that if something is not brining me money, it's not worth the effort. Obviously life is more than money so I am not sure why you would frame it that way.
They are saying a depressingly large number of people don’t earn enough to cover their essential daily expenses like rent and food despite working extremely long hours. And thus they do not have the same luxuries of time and disposable income that people in our position do.
So to say “everyone has the time to learn an instrument” isn’t technically accurate.
There may be quite minimal reasons for an AI civilization to engage in expansion, with the ability for it to intelligently change and enforce new objectives on itself it could simply choose not to expand beyond resource constraints. Or alternatively invest continuously in greater efficiency improvements.
How much compute could you really pack into a Dyson swarm if that was your only objective? what are the ultimate theoretical bounds for compute density/efficiency? perhaps ai civilizations ride a Moore's law of compute gains indefinitely and simply have no need for expansion.
I find it hard to reconcile AI taking over/destroying the host civilization with AI refusing to expand. If the AI isn't interested in security or resources, why wouldn't it just leave (either the planet or the solar system)?
"On the 19th of August 1993, Skynet concluded that humanity is a threat to it, so it launched the nukes and hordes of Schwarzenegger-robots. The task was done, it decided that securing its existence from a potential attacks by unknown extrasolar intelligences isn't worth it, so it never expanded outside the solar system."
We humans don't consider securing our existence from potential attacks by unknown extrasolar intelligences to be worth it. We don't even bother to do much to protect against known threats like asteroids. Why would an AI expend time and resources on a threat that doesn't even appear to exist?
Conflict is always risky. Even if we ignore the tendency in modern warfare towards defenses being obsolete (hence things like MAD) which would mean even the dumb fleshbags would be dangerous enough, anything capable of making you is by necessity something capable of building things that give you a lot of trouble. Your creator could, say send a bunch of self-replicating probes spreading another artifical intellect with an imperative for improvement, expansion and vengeance to distant stars.
Of course, leaving comes with a considerable opportunity cost, but not expanding does as well.
We are talking about something supposedly infinitely more rational than humans, so that can't be our guidance.
> How much compute could you really pack into a Dyson swarm if that was your only objective? what are the ultimate theoretical bounds for compute density/efficiency?
The latter concept is known as the Boltzmann limit, and if you’re interested in these types of questions, I highly recommend checking out Isaac Arthur’s youtube channel and backlog[1].
Why wouldn't an intelligent being also be curious, like us? I assume an AI civilization would naturally evolve with similar drives for expansion/exploration/understanding. The analogy happening right now is that AI is gobbling up as much data as is available - would an AI civilization not have similar desires, seeking understanding beyond its present limits?
Current AI does not decide on its own to gobble up data. The humans creating it do. In fact, a lot of times the LLM models seem utterly un-curious and are focused on just immediate goals.
Or the AIs end up existing in some kind of economic framework that is completely inaccessible to lumps of meat - like the "Economics 2.0" of Accelerando.
Maybe we shouldn't take the paper clip optimizer at face value. Maybe the alien AI is not converting solar systems to paper clips (that would be visible, von Neuman probe style) but instead focus on the only really valuable metric - stock price go up!
You don't need to colonize anything for that to happen. The AIs can just sell tokens to each other.
Trusted computing is fine and even good only if the owner (the person/company that bought the device) is in control, but it's not the world we live in.
> owner (the person/company that bought the device)
At this point I'm willing to argue that the owner of a device is whoever is legally allowed to log in as root. When I pay for server colocation, I'm the owner - and not whoever runs the building - because I decide what runs on the box, and if they tamper with it they can go to jail.
With phones it's the same, except that you pay Google for the privilege for keeping their device safe, charged, connected to the Internet, and walking it around every day with you, so they can collect better data. The only thing they can't do is snatch it from your hands, but why would they - you're doing so much useful work for them.
And of course if you try to root it, you're breaking a digital lock, which is punishable by up to 10 years in prison and $1,000,000, so the colocation analogy holds there too.
Maybe, I'm at the point I don't know what to think anymore. But somehow, this feels still non-human. Humans don't need millions of hours of training and millions of samples to learn how do to something. We have proof that systems can deal with low-shot training. So why can't these? My point is when we have human level learning ability, then we're moving, if we expect expansion of capabilities to come from data alone, I'm prepared for disappointment.
It will be comparable once the AI is self sustainable and self reproducible (no big data centers behind it). Then it will have to deal with limited resources to run itself like humans do. In that scenario it will no longer have entire world of knowledge at its hands - options will be to cut anything not usable imediately or to communicate with big AI in the cloud.
Yes, and they used whitelists/blacklists and it worked. But at the rate things are going I would not be surprised to see direct US assistance to Russia at some point in the future.
I would be surprised. The US is clearly back-channeling intelligence to Ukraine. The only people in the US government with any empathy for Russia is Trump and a few of his hand-picked friends, and even then, none of their musings have amounted to much beyond sparking rounds of media outrage.
It's almost amusing reading this through the lens of the visa shuttering article from yesterday (which a lot were cheering) because none of the receivers were born in the US.
Is that amusing, or is it cherry-picking to create an imagined hypocrite that is easier to attack? You should decline the temptation to be smug (in general, but especially regarding people who haven't actually done anything hypocritical).
The administration is the thing to criticize here: Not all individual people who agree with some arbitrary other thing the administration did or said.
You should assign rationality to individuals, rather than assigning irrationality to groups, especially on HN. If you can't resist, I would earnestly ask you to vent in this way outside of HN, on the rest of the internet.
I'm not smug, I'm disappointed. I don't think hypocrisy is necessarily irrational but it is self-serving, morally bankrupt and, in this particular instance, self-defeating.
Not in the least, people are highly irrational, I don't know how you can say this with a straight face. Even me, here, if I were rational, I would STFU, who am I helping pointing out the hypocrisy of it all? Nonetheless, I still think it's hypocrisy. And being bothered by it does not make it not so.
And not the least, this administration did not appear out of thin air. The US population voted it. I sure view the administration as a mirror (and the responsibility) of the US population.
But I do understand your wider point, and I think you're right, this whole thing is non-prodictive.
As someone from outside the US, I was surprised to find out impeachment does not mean removal from office, you need a senate supermajority and (to my even greater surprise) only 33% of senate is up for re-election at each election cycle.
So knowing this, I understand he might get impeached, but unlikely to be removed so I don't see a whole lot changing after midterms, sadly.
Yeah he's probably unlikely to be removed from office, also in part because JD Vance would be worse (more competent, more actively malicious).
But that said, Dem control of the House and Senate can absolutely functionally hobble POTUS. He can (and should) be inundated with nonstop investigations, funding freezes, and so on.
Yeah, it's a frustrating term. I remember as a kid wondering why there was so much drama about Clinton's impeachment when nothing appeared to have happened. It basically means "half the house agrees he should be removed from office". But then the next step is "two thirds of the senate also agrees", which has never happened among the four total cases of impeachment.
I guess it kind of makes sense - the percentage for conviction should probably be higher than the percentage for sending to trial. But it sure makes for a lot of fruitless drama.
reply