At least this prelude to the apocalypse is quite funny.
Kimi K3 comes public, a new AI model that seems performing quite well, in a way that is at least somewhat comparable to the most advanced “frontier” models available right now.
This is relevant for a couple of reason: it’s Chinese, and it’s open source.
Open source here means a few different things, but mainly you can use it locally in some capacity.
You should see how this is a significant threat to US hyperscalers, because everything they are spending money on might as well be close to worthless.
This being the overall main discourse between those who are AI-savvy and trying to figure out “what happens next.”
The bulk of these conclusions is roughly this: the difference between “frontier” AI models and everything else is quite minimal, especially in practice. It means that for a daily common use the difference between having access to the best AI resources and the average ones is negligible. There’s almost no difference in practice.
But the real threat is that this higher performance gap also gets closed within 3 months, or less. AI models get better, but all AI models get better. Those who spend the most, whether OpenAI, Anthropic, Google or whatever, spend INSANE amount of money for a minuscule advantage. An advantage that gets smaller and more negligible over time. Something with immense costs, to produce an advantage, that vanishes immediately after.
We all know already “the bubble is gonna pop” (and it should have popped already, if it wasn’t for fascist government protection). But the point here is how badly, because these big US companies have literally nothing to sell.
At a basic level AI was a trojan horse that stole the breadth of human knowledge, siphoning it out of public hands and handing it to private companies. They stole everything so that they could “launder” it through AI mastication, and sell it again as their own. So, banally, if your company produces and sells software, you feel the lure of AI, to accelerate that production. Only to realize that you are handing out your own competence, BACK to the AI private company. The AI company appropriates your work, and at that point two things happen. Either your software company shuts down, or is bought. In the meantime “layoffs”, all around the world, in order to support that “capex” (the expenses, the crazy amount of money needed to build data centers). Large capitals that move from people’s salaries, to hardware (mainly).
Why these AI companies spend so much money, money they do not have? Why taking these risks? Because it’s merely a giant ongoing robbery & looting, “take everything you can, as fast as possible”. What AI is doing, in the eyes of these (few, very large) private companies like OpenAI and Anthropic, is to CENTRALIZE knowledge. The more they sell AI services to other, external companies, the more they send trojan horses out to these companies, to steal their knowledge and shut them down. They progressively APPROPRIATE everything there is. The classic capitalist model of starving large majorities of people so that a few rich get richer. The widening of the gap. They started by stealing books, art, code from every corner of the internet, and so erasing the concept of licenses, since no one in politics cared to stop them. And now they use AI as trojan horses to appropriate production everywhere around the world.
The crack in this system is that it would all work only if then these AI companies could protect the giant hoard they accumulated. What the public Chinese models show is that THEY CANNOT. Because AI was built by stealing, now stealing is the norm. So what prevents another AI company to “steal” from you? (it doesn’t work technically like this, but it amounts to it)
If the genesis of AI has been stealing everything there is, as fast as possible, the unintended consequence is that now these AI companies can’t build any wall, any moat, in order to defend their hoard. The giant tank of knowledge they built, by stealing, is leaking everywhere not so much as a flaw, but as an embedded structural feature. They are now realizing they can’t keep anything in, and that they have NOTHING LEFT(*).
(*) They have the hardware in the data centers, that as Cory Doctorow has said so many times, are fleeting too. Because technology supposedly moves so fast that by the time a new data center is complete, it is also OBSOLETE and needs to be torn down and rebuilt. Why do you think TECHNOLOGY is stalling? Why do you think no new GPUs are out? Why do you think the hyperscalers are trying to block access to hardware components like SSDs and RAM? Because of scarcity? Nope, because they are trying to lock in the hardware, since they couldn’t lock in the software. They are trying to merge the hardware companies, lock them to themselves in a co-dependent way, so that they can control the process. They are trying to build their moat by significantly reducing ACCESS, since they cannot control the USE of AI.
If the model is public, if you can run it locally and it works well enough to cover 98% of practical uses without any perceivable difference… What have these big AI companies to sell you, after they invested TRILLIONS of dollars?
If you were precisely in that position, how would you feel other than being scared and shitting your pants?
I introduce you to Dean W. Ball, head of “strategic futures” at OpenAI.
https://xcancel.com/deanwball/status/2078133895766114412
He starts by being direct-honest. Why should he? Because of what I already described. These models are public, therefore there’s no PR gaslighting to be done. Whoever is within this discussion has easily the means to verify these claims, so there’s no room to try and deny the obvious. Doing so you’d lose your leverage because everyone would know you’re lying.
So yeah, these two should be both truths: it’s a “very good model”, and it is “token hungry.”
When he says “it seems on par with the best public models”, he saying including those his own company sells. Notice the specifics, emphasis on Q1 2026. Models seem to evolve quickly, so maybe now he has a better product to sell next, to be a step ahead… See what I already described above. That advantage is eroding quickly.
Why, oh why, China is allowing this?
It does make sense, for a centralized type of government/dictatorship that existed by enforcing censorship everywhere, obsessed with control… why would they allow the “open sourcing of models this good,” why wouldn’t they try to keep a tight control, top-down. It all makes sense just as long you have a comic-book, cartoon-parodistic view of China government. Of course the real world is more complex than this. (not to glaze Chinese government)
(I’m not going to comment in detail on the sprinkled rhetorical-fascist propaganda jargon, merely pointing at “AGI-pilledness”, “decelerations/accelerationists”, this individual seems to be rather deep fried by political propaganda. But in this case you could even say that these “descriptions” still work well enough in practice to signify directly.)
“all this money we’re investing, we are investing FOR YOU!!!”
Cry more tears. He’s complaining that what the Chinese model is doing compromises his own business. Does not like competition. The trick here is by twisting “capex” as a contextual term, outside the level of the single business. ALL capex is reduced, so all investments on AI are reduced = bad.
He’s thinking he’s the center of the world, just like all AI psychopaths. Once you bank all your life and futures on AI, you’re living in hell and permanently committed to its demonic(*) deliverance.
(*) I’m not using “demonic” here as a moral pejorative. AI is demonic IN THEIR OWN accurate views. The part I omitted from the quote implies he sees AI as much more than what’s available today, and believes in true general AGI. In that pure, precise sense of something other than human with metaphysical connotations. It’s literally demonic in the way it exists outside the plane of current knowledge. These techno-fascists see themselves as literal sorcerers to access a new plane of reality.
Here comes A CLASSIC. THE RED SCARE. Communism fearmongering.
Oh my! Full AI communism. Dystopian hellscape. AI needs to be exclusive to the hands of sociopathic fascist billionaires, because if instead you make it public there are SECURITY RISKS.
Security risks that VANISH just as long this powerful and dangerous technology is reserved to the private hands of none other than LEX LUTHOR.
You cannot trust governments! You cannot trust the people! YOU CAN TRUST LEX LUTHOR!
Who will certainly always do what’s in YOUR, public interest. Not his!
Security risks! Dystopian hellscape!
Only the billionaires can save you! Trust them, not China!
And here comes the lobbying through oblique language. (You might not know, but this dude was a Trump official just before quitting and going to work for OpenAI. As these people are moved back and forth to lobby public governments into doing their own private interests.)
“I would guess”, because I’m humble. Reads instead as: I would pressure the Trump Administration to realize that OUR best strategy is to CREATE (not analyze, not detect, but create out of thin air) large amounts of regulatory risks. In order to stifle the competition when it comes to sell OUR services to OUR US companies.
Because without doing this there would be fastidious capitalist competition in the free market, and our clients might as well be choosing to buy the Chinese services rather than ours.
We can’t have that in this free market system, right? Here comes tariffs, blocks, bans, open market manipulation, lobbying, corruption and everything else that we’ve seen happening under the sun, shamelessly, in the last couple of years.
China offers a competitive product that disrupts the US interests? Block them. We’re libertarian = we’re selfish egomaniacs. Rules only apply to others, not ourselves. There are no morals other than winning advantages at all costs.
And we conclude with the icing on the cake of the baseless fearmongering (remember? “it needn’t be well justified”).
Open-weight models make the world a bit more dangerous. But you won’t notice, until you have a nonliving, invisible, dangerous, and infinitely self-replicating agent escaped from a lab.
(and notice how he uses a debunked Covid conspiracy theory to bank on his fascists fans sympathies, who love these types of alt histories)
Just remember a basic truth: when fascists detail apocalyptic scenarios, it’s because it’s what they are planning to enact.