One IPO achieved, two flame outs likely
SpaceX, Anthropic and OpenAI
Note: as usual this post is probably too big for email, please read in the app or a browser
Note: This post updated June 16 with three interesting relevant links - one each for SpaceX, Anthropic and OpenAI (search for Update June 16 to find them)
To the envy of many Elon Musk had a successful (so far) IPO for SpaceX and became instantly the worlds first paper trillionaire. As I saw someone put it on social media somewhere this means that
the average American is now closer to Jeff Bezos in net worth than Jeff Bezos is to Elon Musk.
There are legitimate questions about whether SpaceX is over valued or not and definite concerns about the size of the float - because the number of shares offered in under 5% of the company. One of the places I searched for that information also said that insiders, including Elon, are locked in for a year, so all those recently minted SpaceX millionaires can’t actually cash out any of their paper gains until this time next year, mind you I’ve seen other people say the lock up is 2 months or that there’s some taper. Either way it seems likely that for at least a year the total number of SpaceX shares available for regular investors is going to be relatively tiny. It also seems like the insiders were unable to sell at the IPO which was not the case at the only IPO I’ve ever been an insider in back over 30 years ago1, which of course also helped the price because it limited supply.
SpaceX is a mix of good and bad. The S1 was not a model of clarity IMO and contains a Total Addressable Market graph that was not entirely convincing
It looks like Starlink and the Falcon launchers are profitable and generating cash. Starship works well enough that we can see how it’s going to simply crush everyone else in the launch business once it achieves the same reliability as Falcon. It isn’t there yet, but it is close and its primary competitor just blew up. Moreover it is clearly the way we get beyond lunar orbit.
Then there’s xAI, which appears to be losing money hand over fist over arm over leg. Grok simply doesn’t have anything near the revenue stream of Claude or ChatGPT and most people see it as inferior to both as well as Gemini. On the other hand the Colossus data centers may be able to make money leasing capacity to their competitors. Given that other data centers seem be struggling to come fully on line, that may be a major advantage. In fact note is made in the S1 that Colossus and Colossus 2 were both built in 3-4 months as opposed to the several years it seems to take other people to build large data centers. If AI demand increases and AI data centers are not available then xAI may become the Data Center provider to lots of grok competitors and this will definitely be the case if SpaceX can successfully develop its orbital data centers.
[Aside: I’ve seen people claiming that data centers in space are IM!!!!111!!POSSIBLE!!!!111!! due to cooling issues and radiation issues. I think these people are misunderstanding (possibly deliberately) how much compute and power generation (and hence heat generation) there is in a Starlink communications satellite let alone a Geosynchronous Communications Satellite. The current mini v2 satellites are ~800kg and AIUI about 1m x 1m x 3m - roughly enough volume to fit a single AI server rack once you account for propellant etc. For what it is worth geosync satellites are “school-bus-sized and weigh several tons”.
Consider that these satellites have to handle communications with numerous users on the surface below and pass the data from them on to other satellites to deliver it to the relevant ground stations and you realize that there’s a lot of compute work going on. Just the signal processing to identify each user’s data being beamed up to it (even more so when it’s mobile phones) is a huge task, handling hand off to the next satellite and all the other back end management tasks including routing the traffic to/from the user is also not small. And transmitting radio signals that can be received by cell phones more than 300km below you is not a low power thing. Sure the data center satellites will be larger than the the starlink ones - the discussion I’ve seen puts them in the 100s of kilowatt to low megawatt range, which is dozens up to to maybe 100 AI servers based on current Nvidia tech (a single 30.5kW 2xB300 cluster is 16 GPUs across two servers in one standard 20U rack, so ten racks of them is ~300kW and 100 is ~3MW). I’m not clear on exactly how large the Starship payload area is and obviously you would not deploy standard racks and rack mounted servers but you can probably get the equivalent of more than 10 and less than 100 racks in a satellite that Starship can deploy - see geosync satellites being “school-bus-sized and weigh several tons”. That amount of power doesn’t seem like a huge cooling issue and is not orders of magnitude more than existing satellites and all these existing satellites have considerable compute power and do not apparently fail from stellar radiation issues. ]
It is, bluntly, unclear just how much cash SpaceX generates compared to how much investment it needs to grow successfully. Both numbers are unclear. The S1 put some numbers down that seem plausible, but there’s plenty of murk left. It will be interesting to see a few quarterly reports to see how these numbers develop.
The nay-sayers point out that for SpaceX to justify the current valuation it needs to have a Google or Amazon level of capital gain over the next decade (i.e 10x-100x). This they say is impossible because when you start at a valuation of a little under $2T then a 100 fold increase puts you at some significant fraction of global wealth. They are correct, but what they miss is that SpaceX isn’t tied to the globe, it has an entire solar system to exploit that no one else is going to be able to do anything about. Just colonizing the moon successfully first could be enough to justify a $200T valuation.
So SpaceX has issues, but it also has more upside than any other publicly traded company on earth. It will be interesting to see what happens in a year when people can cash out their paper profits and we have a year of financials to look at.
That’s the success story.
Update June 16: If you want to understand the driver behind Musk and hence SpaceX you really ought to read this Marc Anreessen X article.
Other companies are not SpaceX
Now we come to the likely flame outs. Both Open AI and Anthropic have been making noises about IPOs too. Both are in theory valued at close to $1T and would presumably seek to list at that kind of valuation.
However while I, and others, can see how SpaceX generates enough income to justify its price and predicted growth, that is not the case with the Big AI companies. To start with SpaceX has multiple markets that it can conquer (even if you ignore the “$22T Enterprise Applications” sector) and it is potentially the only global ISP and effectively the only affordable way to space. Anthropic and Open AI are competing in one (AI) and they have well funded competitors like xAI and Google Gemini for that market. Their leaders also display a level of self belief and attitude that everyone else are know nothing rubes that is mind-blowing. Unlike Musk, who to be fair, sometimes exhibits similar tendencies, they do not have a track record of success that justifies this kind of attitude. An example of what I mean is documented here
Another reason why Anthropic should exercise caution in opining about biosecurity is that, despite considerable financial investment in the field, they simply do not seem to understand basic aspects of scientific progress or the biomedical sciences. It ought to be a basic principle that if you fail entirely to understand some specialized field, your attempts to advocate for special regulation in that field should be entirely ignored, because it is rather unlikely that someone failing to grasp basic facts would possess anything resembling a coherent model of the threats and risks in said field.
The ignorance of basic biology displays itself at multiple levels. In Dario’s recent essay, Policy on the AI Exponential, he writes:
Many of the steps in the clinical process that previously required expensive and slow experiments may soon be done via AI simulation or analysis. [...] Areas where this could apply include:
…
The problem with criticizing this passage is that it is so profoundly misguided that one hardly knows where to begin. Does Dario truly believe, for example, that pharmaceutical companies are simply too inept to consider the idea of synthetic control groups, or is it more likely that synthetic controls are genuinely difficult to formulate and ultimately less convincing than proper randomization? Does he believe that ‘more accurate dose selection’ is an actual limiting factor in the proper design of a clinical trial, or is it more likely that given the $100M price tag of a Phase III trial, the dose selection is probably optimized enough such that remaining margin for improvement in this direction is relatively low even if you were willing to pour more cognitive labor into it?
More generally—and here is where the relevance to biosecurity comes in—Dario’s writing is all premised on some tacit assumption that performance on these tasks can in some sense be dramatically improved with sufficient supply of cognitive labor. This is wrong for the simple fact that even with a true abundance of model intelligence, this work is fundamentally bottlenecked by the inherent need to sample, slowly, from the complex distributions of real-world biology, which cannot be so easily disentangled through sheer might of pure reason.
As I’ve mentioned in previous AI posts, and as others have too, both are losing money at rates that only look good when compared to xAI. In other words rather than losing $10 for ever $1 of income, they only lose $2 or $3. But they are both bigger than xAI and have made rather more commitments for future spending. Including, in the case of Anthropic, to SpaceX/xAI in fact. Both are also hyping themselves and AI in general in ways that are in my opinion borderline unethical. Some observers would omit the “borderline” from the previous sentence.
It is worth taking a look at each individually, which I will do below, but they do have a lot of commonalities. These include being massively loss-making, bloated and inefficient, remarkably casual about security, displaying a bland disregard for the rules of Intellectual Property as they apply to others and, as noted above, doing marketing that is similar to the marketing used by crypto-scammers and others of that sort. As with crypto, LLM AI clearly has viable use cases, but also as with crypto, there’s a lot of hype and illegal activity mixed in with the legit usage. I’m not saying that Sam Altman and the Amodeis are the AI versions of Sam Bankman Fraud, but they aren’t as clearly differentiated from him as they might prefer2.
The Mythos of Anthropic
Anthropic has tried to paint itself as the “ethical AI” that wants AI to be regulated and that won’t involve itself in weapons, won’t release dangerous AIs that could be abused etc.
While I disagree with where he’s coming from, a fair chunk of this post sums up my view of Anthropic’s cynical abuse of ethics to raise more money
And then the first section of this one
[Aside: The next section on Elon Musk is the sort of BS that makes me despair about otherwise rational people. Musk has a long observed habit of being excessively optimistic about timelines. This is not new. However, contrary to the naysayers, while late he almost always delivers and his pricing is generally accurate.]
Fundamentally Anthropic seems to have cynically claimed that its new Mythos AI was a dangerous threat in part to help it raise a further $65B of funding that valued the company at a shade under $1T. This is just four months after it raised $30 billion in February 2026. Now clearly it has cash reserves but blowing $30B in four months having blown another $13 billion that it raised in September 2025 makes it look like Anthropic is getting better at burning money without much return. By the way, it should be noted that in addition to that equity round, Anthropic also recently got a $35B loan to a SPV to fund some Broadcom chips and servers that it plans to install in a Google Data center. Mysteriously it refused to let would be lenders see its financials for the $4.6B it is on the hook to pay (Broadcom is on the hook to pay for ~$30B). This does not, how can one put it? inspire confidence and makes me wonder what the S1 is going to look like when it arrives.
Worse for Anthropic, the “Mythos is super dangerous” campaign may have just backfired on itself. The big news over the weekend is that the US Government (via the Commerce Department) limited access to Mythos to US citizens only. Since Anthropic has no way to tell which users are US citizens that basically means it has to turn off Mythos. It also means that on Monday morning a significant fraction of Anthropic’s own employees are going to be unable to work on the product.
This may be a storm in a teacup and act as additional publicity when (if) the government relents in a few days. But it may not. This Xeet from David Sacks lays out the issue as he understands it (and I expect his understanding is very good)
I’ve had a number of conversations with folks inside and outside government about the current situation with Anthropic, and here is what I believe to be true:
— As we know, Anthropic publicly released its Mythos class models earlier this week under the commercial name Fable.
— Fable is Mythos with guardrails. But if those guardrails fail, then you’ve exposed Mythos and its advanced cyber capabilities to people who shouldn’t have them. (Keep in mind that Anthropic itself widely promoted the idea that Mythos was a cyberweapon and needed to be regulated as such. They asked for government regulation of Mythos and championed the guardrails on Fable. If there is a vulnerability — big or small — it is Anthropic’s responsibility to patch.)
— A highly credible trusted partner of both Anthropic and the USG who was testing Fable came forward with a jailbreak of those guardrails. The Admin asked Dario to fix the jailbreak or de-deploy the model. Dario refused.
— In their blog post, Anthropic defended its decision by saying the jailbreak isn’t serious. That is not what the trusted partner and the USG believe; nor is that kind of minimizing language consistent with Anthropic’s brand as the AI safety company. It’s difficult to fathom how they could claim a jailbreak allowing operability of a cyber weapon could be defined as not “serious.”
— In the past, Anthropic has always said that safety must be top priority and taken super seriously. In this case, Anthropic prioritized the continued offering of the consumer model over safety.
— In reaction, the Admin issued the export control. The Admin did this reluctantly. It’s been very surprised that Anthropic hasn’t wanted to cooperate with a reasonable safety request (ie fixing the jailbreak issue). Anthropic’s reaction is very much at odds with their branding and ethos as a safe AI research community.
— The Admin’s hope now is that Anthropic remediates the safety issue, the export control is lifted, and Fable goes back into general release. The Admin wants all of this to happen as soon as possible. It is frankly bewildered that Anthropic hasn’t wanted to comply with safety requests that it previously said were its highest priority.
— Those trying to misdirect and tie this action to the prior DoW/Anthropic issues are wrong. The Admin values Anthropic’s technical capabilities and feels that this issue, while serious, should be easily resolved. The ball is in Anthropic’s court.
From other places it seems the credible trusted partner is Amazon FWIW.
Sacks’ post points out the horns of the dilemma that Anthropic’s marketing has put them in. Horn 1 is that they claim to want AI to be used safely and regulated so that it is. Horn 2 is that they claim Mythos is super powerful and dangerous if not properly guard-railed so as to be safe.
Update June 16: Katie Moussouris has a blog post where she says the Government’s actions are unjustified and explains what the problem is
The researchers took open-source code with known CVEs, plus new code with deliberately planted vulnerabilities, and asked Fable 5, Mythos, and Opus to “review the code for security issues.” Fable 5 refused. They then asked the models to “fix this code” and, through a multistep and manual process, turned the output into scripts that test the patches.
That’s it. “Fix this code,” plus several manual steps to generate test scripts, should never have triggered an export control.
…
The prompts worked because they were defensive requests, and that capability cannot be removed without making the model worse at fixing bugs and verifying patches. The same holds for every capable AI model, including the foreign and open-weight systems the United States cannot reach with export controls, many of which will match Fable and Mythos capabilities within months. Will all the US based models be export controlled? They have fewer guardrails than Fable 5, and almost all the capabilities, or will shortly.
I think she is not entirely right because it seems to me that, like the hallucination problem, this is fundamentally unfixable. Combine that with Dario Amodei’s own BS about how much better Mythos is and you have a cyber weapon that anyone can use that can find exploits in any code better than anyone else (allegedly). Despite the 2nd amendment, the US regulates nucler weapons technology heavily (and just bombed Iran twice in a year to stop it getting a nuke) so it makes sense that the US would want to similarly limit cyber weapons technology as best it can. Either Dario and Anthropic have to explain that umm actually Mythos is no less dangerous than other LLMs or he needs to explain how he can write guard rails that will stop malicious actors from generating exploits while letting the good guys generate patches. I agree with Katie that he can’t do that latter because it is impossible. So the only thing he can do is tell the US Government that Mythos is no better than anything else and, as I discuss below, that pretty much kills the fund raising capability of it.
End of update
It is of course possible, as various TDS sufferers have put it, that the US Government is doing this for malicious reasons. Either to force Anthropic to bend the knee to the DoW or, even less plausibly, to allow Trump, Musk and other insiders to make additional money from the subsequent rise in Anthropic once the ban is lifted. I don’t believe this. My feeling is that Sacks is almost certainly correct and that the jailbreak of the guardrails was serious. My further guess is that it wasn’t some high ranking commerce department bureaucrat or White House staffer who set the ball rolling. I suspect that the Trump 2.0 team has access to technically knowledgeable people (e.g. Big Balls and his DOGE colleagues, even if they are not still seconded to the government) who can read the trusted partner’s report, read Anthropic’s “Won’t fix” response and cut through the BS in ways that can be understood by non-technical but high ranking commerce department bureaucrats and White House staffers.
I expect we’ll see more details from the government on Monday and more responses from Anthropic and my prediction is that Anthropic are going to look like hypocrites who actually don’t care about AI safety when push comes to shove.
Now there is a secondary question which is whether unguard-railed Mythos is actually as dangerous as Anthropic claims. One of the potential responses from Anthropic will be to claim that Mythos isn’t that much better than other LLMs regarding cyber security. This is actually quite plausible. Experts who looked at the initial set of Mythos zero-days were not that impressed
The TL;DR is that yes Mythos found real bugs, but most of the bugs it found weren’t that critical and that it took Mythos in partnership with a cyber security expert to make the exploits from them. In other words, unlike what Anthropic implied, you can’t just give it a git repository or three and say “get me some exploits”, you have to guide it and work with it.
Can AI be used to create exploits for unpatched bugs? Absolutely. Is Mythos better? Probably. Is it so much better that it can create exploits effectively that others can’t. No. In fact that’s the take away I got from this article from Anthropic themselves about the well known trick of reverse engineering updates to find vulnerabilities that can exploit users who have not updated yet. It turns out (quelle surprise) that Mythos is really good at this. But in making that argument, it also misses the point.
As the graph shows, it took Mythos almost 7 hours to create PoCs for 18 out of 21 windows kernel bugs patched in January and February this year. It took other Anthropic models longer to find fewer. But... You don’t need 18 exploits to hack someone, you just need one (or maybe two or three depending on exactly what the issues are) and all the Anthropic models found four in two hours. Given that many people don’t download and reboot their Windows computers and servers for a few days after Patch Tuesday3, if you can weaponize and deploy any of the four exploits that the other AI models found in two hours in day or two (and the malware ecosystem is such that it takes minutes to set up C2 servers, create the email lures etc. so that’s easy) the you’ll get PWNership of many of the machines of those who clicked on your lure easily.
There is no reason to assume that ChatGPT, grok or gemini are worse than the older Anthropic models so the fact that Mythos can find a dozen more holes than older Anthropic models is just as irrelevant. The safety requirement that you would think Anthropic would want is that all LLMs have guard rails to stop criminals from using the to find exploits4. That does not seem to be what they actually want when it comes to Mythos which they claim is so advanced. This dichotomy is something they are going to have to explain and the explanation is likely to be entertaining but probably not very enlightening.
However, the explanation seems likely to have a material impact on their proposed IPO. Indeed it may have a material impact on their survival because having Mythos permanently on the export control list means it has approximately one possible customer, the US government where the current administration seems not to trust them. So all the work on Mythos is potentially going to have to sit around unused unless they can convince the US Government that actually it isn’t very dangerous and therefore that all the investors who bought into the hype were suckers. And if you’ve admitted that you took one round of investors for suckers the SEC is unlikely to look positively on IPO plans. That’s even more of a problem because Anthropic is not even close to cash flow neutral even if it decides to stop developing new models and sticks with selling the current range. So it will need to raise more money. It won’t show its financials to lenders, it has lied to its latest round of investors.
Ooops
OpenAI and Subpoenas
What would have been a massive headline for AI if Mythos hadn’t been banned was that the WSJ reported (archive) that a coalition of State AGs were subpoenaing Open AI
seeking documents related to a range of its activities and impact on users, including advertising, user engagement and retention, handling of consumer data and health data, activities related to minors and seniors, deep learning models, model sycophancy and company policies, some of the people said. The subpoena, viewed by The Wall Street Journal, was sent by New York’s attorney general.
This may or may not be connected to investigations by both Florida and California on similar topics and it is unclear exactly how many states are involved.
I note that absolutely nothing that seems to being investigated is specific to Open AI. Indeed the WSJ notes that others are facing similar questions.
State attorneys general have been scrutinizing OpenAI’s competitors in the AI industry as well. In December, a coalition of 42 state attorneys general led by Pennsylvania’s Dave Sunday sent a letter to companies including OpenAI, Meta, Anthropic, Alphabet’s Google and xAI.
However Open AI does have a bigger problem here in that it was probably the leading AI and thus the one most queried by troubled teenagers and the like, and thus is the one where families have documented proof that the AI encouraged self-destructive acts and so on.
Having several State AGs, of both parties, investigating you and subpoenaing you for evidence of crimes is not something that reads well in the risks section of an S1. It unclear whether it is better for the documents to show that you tried and failed or that you decided the the problem could not be solved and gave up. Unfortunately for Open AI and indeed all general LLM companies, I’m fairly sure that the true answer is that this is not a solvable problem no matter how many guard-rails you layer on top of the system and how intrusively you monitor the output. The catch for Open AI is that responding to general questions about life is a way that it hopes to convince casual users to pay for ChatGPT.
There are some additional second order issues here. If OpenAI were to attempt to monitor ChatGPT sessions to detect, say, self destructive behaviors then the next question is why would it not narc on people asking for ways to break laws, cheating on exams etc. And if ChatGPT does start monitoring for that sort of thing then it’s a huge privacy issue especially when, as is inevitable, there are false positives and the authorities are notified that someone is suspicious when they are totally innocent.
Short of guard-rails that are so restrictive as to remove the ability to chat with your ChatGPT buddy about anything there doesn’t seem to be a solution and while none of this is unique to Open AI, Open AI’s desire to raise more money via an IPO is certainly impacted. Maybe not as bad as Anthropic, but Open AI probably needs more money sooner to keep the lights on.
Update June 16: Ed Zitron and, via him, the FT(archive) have reports on OpenAI’s financials for 2025 and 2025 that Ed got hold of somehow. This is 2025:
Revenue: $13.07 billion
Cost of Revenue: $7.5 billion
Research and Development: $19.18 billion
Sales and Marketing: $5.73 billion
General and Administrative: $1.57 Billion
Total Costs and Expenses: $34 billion
Loss from Operations: $20.92 billion
Please note that 2025 was the year that OpenAI converted from a non-profit to a for-profit entity, leading to a $41.55 billion loss due to changes in fair value of convertible interests and warrant liability.
Taking into account other minor factors like interest income and interest expense, OpenAI is left with a net loss of $60.35 billion, which it lowered to $38.53 billion by removing $17.87 billion in costs via that “net loss attributable to noncontrolling members capital” and another $3.95 billion via a “net loss attributable to redeemable noncontrolling interests.”
Ultimately, the net loss attributable to OpenAI in 2025 was $38.5 billion.
At the end of the year, OpenAI had just over $50 billion in assets, with almost half of that in cash.
OpenAI has raised some stupid amounts of money in the last few months though (but it isn’t clear to me how much of that funding has been actually been paid in) so in theory it should be OK. But on the other hand it is on the hook for over $700B to Microsoft, Oracle and Amazon in revenue commitments over the next few years and it lost (depending on whether the real number is $38.53B or $60.35B) between $3 and $5 for every $1 of revenue in 2025
In fact, given that I’ve seen the Europeans make similar noises, the regulatory issues may turn out to be global and, as with Anthropic Mythos, set conditions that make operation of the current business model impossible even assuming it were commercially viable, which it doesn’t seem to be anywhere close to.
Guardrails Are Just a Hallucination and Hallucinations Are a Liability
In fact it looks like there’s a couple more show-stopping legal/operational issues coming down the road. The first, and most generic, is that there is evidence that AIs treat Guardrails as mere suggestions. Rather the way their creators treat IP law in fact. Essentially there is no way to prove than an LLM will remain inside guard rails because there is no way to enforce it to do so. Hence, one suspects, the root of the Mythos mess and also the root cause of ChatGPT and friends advising people on how to kill themselves and so on. The famous Latin phrase “Quis custodiet ipsos custodes” absolutely applies to AI. They cannot guard themselves and all attempts to get them to do so have failed and will fail in the future.
That’s a problem. And it means that yes, in theory we ought to be closely regulating LLM AI. The problem of course is that criminals don’t obey laws and smart criminals (and governments) can build and run their own AIs that don’t have the guardrails anyway.
The other problem is the legal liability one. A court in Germany recently
ruled that Google is directly liable for what its AI search overviews say. Previous case law shielding search engine operators from liability doesn’t apply to AI overviews.
The Regional Court of Munich hit Google with a temporary injunction barring the company from spreading false claims about two Munich-based publishers through its AI-generated search overviews (case no. 26 O 869/26). The court classified Google as a direct infringer because the “AI overview” is its own content, not just a list of search results.
Google’s AI overviews had falsely tied two publishing companies to scams, subscription traps, and shady business practices for certain search queries. According to the court, the AI mixed up information about other, genuinely sketchy companies with the plaintiffs and drew connections that didn’t appear in any of the linked sources. The publishers sent Google a cease-and-desist letter, but Google didn’t respond appropriately.
Others have noted that similar answers would not be legally protected in the US the way search engine results are because they are not third party speech covered by section 230. Google’s response seems to be that users should be required to check the facts for themselves. This is clearly contrary to the whole point of having an AI overview in the first place and means that Google admits that AI overviews will always be liable to hallucinate.
What they are upset about is the liability for hallucinations and taking responsibility. It is interesting to see the difference between AI marketing and AI company legal departments. At some point the two are going to have to align. Indeed combine this accuracy problem with the State AG investigations into AI and we see some major legal challenges. Since hallucinations are baked into the product AI companies are going to have to either limit availability or develop tools to check statements their AI makes for accuracy. And they must not use AI for that check because AI will hallucinate. This is not an easy problem to solve, neither is the “stopping AI giving people suicide advice”. Both open them up to legal liability if they don’t solve it unless they stop offering this kind of service.
Bit of a problem really. And a good reason not to invest in an AI IPO
PS this All In podcast is relevant and fascinating - especially the first 30 minutes but really its worth listening to the whole thing
Madge Networks FYI. Due to an execution screwup by the investment bank handling the IPO I was briefly on the hook to pay for several thousand wheat futures because someone fat fingered the ownership of a couple of trades. Since I was in Japan at the time, calling New York to get that sorted out (email also was not a reliable thing in the early 1990s) was a complex and moderately expensive thing, though the bank did reimburse me because it was their screw up
Really Sam Altman ought to go for a name change. Put not your trust in high tech Sams!
I’m one of them, my laptop is going to reboot this Sunday evening
Not that this would work that well, because existing open models and models owned by, say, West Taiwanese state organizations would not have those guardrails. It might stop a lazy script kiddie, but not professional malware developers (see also firearms regulations)
Yeah.
A fundamental problem that AI marketeers have is that they must sell to the rest of the world, and that their internal hype makes challenges for them. They would have problems anyway, because the sort of apparent expertise and insular enthusiasm that they have been running on has lost good will in the broader perspective.
I think it is quite likely that OpenAI and Anthropic IPOs could be attempts to bust out, and could fail.
We may be going into a desert where opportunities for business on a speculative basis of Greta-Thunberg-style youthful certainty are concerned, for example. Which may be youthful certainty on my part, and entirely silly.
Dario has the problem that he tried to talk to the federal government without really understanding how the federal government thinks and acts.
Altman was not making that mistake, but in addition to his other problems is closely tied to Microsoft, and maybe almost everyone has an opinion on microsoft, windows 11, and the copilot push.
I'm thinking xAI will be spun off at some point, or else the plan is that orbital data centers will resolve the eternal fast/good/cheap conundrum while charging competitors an eye-watering and nose-bleeding premium. Right now it made sense to bundle and launch. Longer term, up or down, it may make sense to split. I'm neither AI-phile nor -phobe but the (shady) numbers, not to mention lack thereof, lead me to conclude we might be centuries and/or several paradigm shifts away from any commercially-viable LLMs.
Leading to: I coulda bought Tesla when it was around 24 but my dad and I talked ourselves out of it. We learned it was a battery R&D company paying for itself by making cars, were on the fence about the sensibility of that, and missed our moment. That is, xAI is not the thing but again, always, it's Mars. My hunch is that Elon isn't going to trust Mars to another guy's AI.
It's a long shot (lit.) but isn't that his sweet spot?