So the the operating expense was greater than their revenue from operations by about 2, but it seems like they’re minimizing it by hiding the cost of some of the compute inside marketing and training costs. This is something that a few AI companies in China have been caught doing to make it seem like they’re doing better than they are. So they could be incinerating money at an even faster rate than they just admitted.
I mean, it’s built on chrome, like opera and edge, so if Google is on the war path they’re probably not safe long term.
I suppose they could hard fork the web engine, but, that means they gotta develop a web engine which is a whole other can of worms.
If Google is out for blood on this, only browsers built on WebKit or Gecko would be safe.
I mean. I’d argue that the group that more frequently questions its positions and beliefs based on new evidence without killing a significant portion of a country, is a safer source for interpreting the world than the group that killed 40% of Germany when one assumption, regarding the authority of the bishop of Rome, was in dispute for the first time in several centuries.
I think a lot of people will brush it off what ever happens with SpaceX because Xai is the weakest of the big model providers. I don’t think we’ll even get to the openAI IPO. I think the S1 or IPO from anthropic will be the real kicker, because they’re the darling at the moment. The one that seems to have the most momentum and business customers.
If their S1 turns out to be utter dog water and/or their IPO flops, that’s when the real panic comes. When people say “oh crap, these GPU data centers don’t have customers that can make money”, and that narrative will hit all the people building data center and all the people selling hardware to them. Amazon, Google, Microsoft, Nvidia, that’s like… what, 25% of the S&P 500? That seems like a big enough hit to catch a lot of other stuff in the blast radius.
i looked up the story I had read a while back, and sundar was not the head of ads, but had just become the ceo after having been on the board of directors, and the issue was kicked up to him when Ben Gomes refused implement changes that Prabhakar Raghavan of ads wanted. later Gomes left the company and was replaced by Raghavan as head of search.
This all coming from internal emails and memos that were released as part of the anti trust case summarized in this piece which does a very good job of sourcing individual claims by pointing to specific emails and memos, although, the author is very… passionate in their coverage.
Part of me feels like google is comfortable allowing their traditional search experience to languish because it makes ai more necessary
Oh, it’s so much worse than that, Google intentionally made it worse around 2019 so that people would do multiple searches and scroll to second pages, thus increasing the amount of ad impressions and user time spent on the site. There were several email back and forth between the head of search and head of advertising, with the head of search adamantly refusing to implement the changes. Eventually he ended up leaving despite having been at the company from the beginning, due to this disagreement. The head of advertising during this? He’s the head of search now.
Replacing search altogether with AI summarization is just a continuation of that, instead of delaying customers going to other sites, prevent them from going to them all together.
To clarify why the S1s are so important here. There are a ton of laws that prevent companies that have their shares bought and sold on public stock markets from outright lying about the state of their business to the public. This is to ensure that companies can’t just lie to push up share prices.
Since the most important model providers are currently private, they have not been bound by these laws and have basically been allowed to say what ever they want. With them going public, and the release of the S1s, they need to publicly display accurate financials, not just vague and un verifiable numbers.
The question of “is this just a fad or is this a revolutionary technology that will reshape the economy” is easily answered if we have accurate accounting of the costs to run these business.
This stuff has been there. Just not much reporting or investigation has been done on it, instead a lot of news outlets focused on interviews with the heads of these companies and thoughtlessly reiterated press releases. Those who have criticized it have mainly just done so in ways that play in to the narrative convenient for the industry “but what if it’s TOO good?”.
The failure of tech journalism on this topic has been frankly catastrophic.
To break even on operating expenses, not even counting debt payments, depreciated capital value, or future recapitalization costs.
Just gotta configure and tweak until it gives outputs you find indistinguishable from correct. Just gotta train it to gaslight you properly. Come on don’t you want to be given and endless stream of stuff that looks correct?
The data center builders and operators are running behind on basically every data center project, massive amounts of Blackwell are just sitting in warehouses right now because they loose money the moment they’re plugged in. New IT load for the operators isn’t even a third of the capacity of the chips sold to these companies. Most of the companies that got switched over to token billing by the model providers are pulling back and freaking out over a yearly “AI budget” annihilated in a quarter. The model providers wouldn’t have swapped to token based billing for enterprise clients if they didn’t have to. They’ve been pumping demand with hype and by selling at a fraction of operational costs. Reporting non-gap EBITA to hide what a mess the financials are.
The SpaceX S1 alone shows how insanely cash incinerating it all is. Can’t wait to see Anthropic and OpeAIs S1s. It’s gonna be hilarious.
The fact that they tried to get the rules changed so they could get listed on index’s as soon as they IPOd means they’re out of willing creditors and out of cash. They were attempting to dump equity on to 401Ks to pay out the private bag-holders who have been funding them to sell model access at massively discounted rates.
Even if people host models them selves, the era of constant slop deluge is over simply because all the players giving away access for essentially free are about to go belly up or pivot away from it to other business models.
Oh, you didn’t want to be disoriented by all the apps flying apart in every direction when ever you wanted to use the task bar? Oh you wanted a system tray not hidden behind a menu?
Oh, well you can just use a plug in … just pray we don’t update and break all the plug ins anytime soon.
Ok, but… what if the bank asks where you got the gold? Or they ask if you have proof you came by it legally, and then the IRS notices you have a bunch of outstanding loans collateralized with an asset they have no record of you obtaining or paying taxes on the income used to obtain it.
Like… with the cash, you just… spend it on stuff. If you need to pay for something not in cash, then just open a cash based business and overstate your sales.
Or you could… just… list those things as unspecified income and pay taxes on it and not worry about slipping up and going to jail.
It must remain unharmed
I don’t think these companies are worth the value of the paper that their charters are written on. I don’t think many of them are going to last much longer, but he might actually be on to something here in terms of diffusing a massive financial fraud that’s about to happen in broad daylight.
The common wisdom right now is to just “put your money in an index fund, it’s safer and outperforms actively managed portfolios.” Which is to say, a fund that just buys a little bit of everything from a given list, rather than trying to pick stocks that someone thinks will do well, it creates a very diversified portfolio that is protected from anyone company fucking up by having the value spread over as many things as possible. Because many people just put their money in index funds now, getting listed in an index kind of guarantees that lots of people will be buying the shares consistently and thus consistently causing the price to rise over time.
To get in to these lists, normally, a company has to have been public for a while, generally about a year, and show profitability for a prolonged time. These rules have recently been changed though.
A bunch of the AI companies are doing initial public offerings (IPO meaning putting their shares on the public markets for the first time) in the next few months. And stating absolutely insane valuations. Because of the rule changes, they’re basically all getting immediately listed in index funds. And since they’re all targeting insane valuations, they’re going to automatically suck up a bunch of retirement money by default.
I’m not sure about anthropic and openAI on this next part, but SpaceX (which just “bought” Xai), is only going public with about 5% of it’s shares, so theoretically they can just trickle more shares on to the market to get bought up by index funds, and because supply of actual shares is artificially constrained, it will lead to massive overvaluation of those shares, taking up a disproportionate amount of money going in to index funds.
It’s actually fucking criminal that this is being allowed to happen, but because the rules for index funds and IPOs are set by financial institutions and stock exchanges, with very limited oversight by the government, they can just do this.
If the government were to take these 50% shares, it would kind of throw a wrench in the plans, since it would give the government the ability to sell those shares on to the market and stop the over valuation that allows them to take disproportionately from the index funds.
Asahi is still getting a lot of work and development done, but since they’re reverse engineering everything without proper documentation from Apple, they basically get kicked back to square one on any component that Apple makes big changes too with each new generation of chip. They’re more focused on polishing off support on older chips and keeping what they have up to date rather than jumping on every new chip release and leaving stuff unfinished on the older ones
So right now M1 and M2 Macs are fully supported. M3s are mostly working but very much a WIP, and M4 hasn’t really gone anywhere yet.
I’d say that if you find the advantages of the M chips really compelling, but want to stay on Linux, an M2 running Asahi is still very competitive, and the work to support the M3s is chugging along.
If you want the newest hardware possible, then yah you’re probably better off with an AMD or Intel based computer. There are also some other ARM based desktops out there, and Linux on arm is a thing, I have no idea what’s going on over there though.
the leadership at a lot of companies have a very poor read on public sentiment, kind of strange given how much data they collect and how much they like to talk about how good they are at using that data.
And a lot of high level leadership at collages run in the same circles are executives at big companies. These speech events are sort of a benefit for both sides, the leadership at the collage gets to advertise what a good job they’re doing that they were able to get someone so influential to speak, and the speaker gets a sudo-academic platform to state their ideas and an ego boost from the huge in person captive audience.
A lot of them just kind of write off the discontent they see as “a vocal minority”, so when mass confronted with actual public sentiment, i do think it kind of blind sides them.
Yah, given that “training models” doesn’t stop when the model is finished and released. Like, a released model needs to be continuously tweaked to keep it up to date or to deal with problems that have occurred. Even if that’s not literally tokens used by customers, it is compute being used to provide service to customers.
And that’s just assuming that they’re not just hiding some compute costs used to service customer demand inside the R&D budget. “Oh, you see, this pool of customers are being served with an experimental version, so any compute here is actually R&D, any API fees or subscription payments made by them of course get counted towards normal revenue.”