Editor’s Note: Due to the length of this piece, you may need to click a button to read the whole thing in your email. I also made a mathematical error in calculating the cost of Grok AI - this has been fixed.
Last week, dressed like a destitute bombardier, Elon Musk delivered a stark message to advertisers: don’t want to advertise on X? Go fuck yourself.
This was the climax of Musk’s two-week-long tailspin that began with him publicly endorsing an anti-semitic conspiracy theory, leading to companies like Disney, Apple, IBM, and Warner Brothers Discovery pulling their advertisements from the already-suffering platform, leading to Musk suing non-profit group Media Matters for defamation over its report that showed advertisers’ posts alongside neo-Nazi and white nationalist content. Musk’s argument, for the most part, appears to be that Media Matters “tricked” the website into showing the ads by creating new accounts and refreshing the page to see what comes up, otherwise known as “using the website that used to be called Twitter.”
Musk’s anger, as ever directed anywhere other than at himself, is likely because the majority of advertising spend happens in November and December, and as Marketplace notes, Twitter reported $1.57 billion in revenue in Q4 2021 — with 90% of it coming from advertising. Twitter’s advertising revenue has declined every single month since Musk acquired the site, and its own internal projections show an estimated $75 million loss in revenue during the holiday season, with big advertisers like Netflix, Airbnb, and subsidiaries of Microsoft costing them nearly $10 million in revenue alone.
What’s more astonishing is how long it took advertisers to finally walk away. Musk has a long history of racism, sent tens of thousands of dollars to right-wing lunatics through Twitter’s creator program, and unbanned an account that had been kicked off the site for posting child sex abuse imagery. Advertisers didn’t seem to blink at the increase in hate speech just under a month after Musk acquired the site in 2022, nor did big brands like Walmart, IBM, or Apple disengage despite Fortune 500 CMOs’ concerns about him “perpetuating racism.” They didn’t blink when he banned journalists, sued multiple non-profits, got sued by the SEC, unbanned multiple neo Nazis and bigots, or threatened to sue the Anti-Defamation League. While advertising had begun to crater on the site, it has become very obvious how many large companies were totally tolerant of a man who has espoused conspiracy theories, helped harass journalists, and the fact that he has a fairly broad history of outright misogyny.
Someone less cynical might see this as a compounding of problems over the last year of Musk’s tenure, as advertisers repeatedly tried to kid themselves that these were flash-in-the-pan moments from a troubled genius rather than a continuum of bigoted actions. Musk has never had to reckon with society’s consequences before, escaping blame for calling a hero saving children a pedophile, outright manipulating financial markets, or creating an open-air beta test on the world’s roads of Tesla’s autopilot system.
It’s because, at least before this point, his loathsome nature had never really inconvenienced the rich and powerful. Musk’s previous statements, though disgraceful in a way that many people could see, were the kinds of things that one could wave away if they didn’t really care, calling Musk a “firebrand” that was “hard to pin down,” knowing that the average person likely doesn’t keep up with every single thing he says. Advertisers had previously “paused” their spend (and that remains the language they use), but CNN reports that brands have begun to entirely disengage with the platform, including Disney (who Musk specifically referred to after saying “go fuck yourself”), Paramount, Sony Pictures and Universal, who haven’t posted in nearly two weeks. A report by the New York Times said that advertisers didn’t plan to return to spending on the site “anytime soon,” with half a dozen marketing agencies saying that their clients were “standing firm against advertising on X.” It also reported that over 200 advertisers have halted their spending.
After months of posting gleefully empty statements about nothing as Musk continued to swerve as hard right as possible, Twitter CEO Linda Yaccarino has finally accepted her fate and wholeheartedly embraced her right-wing heritage despite marketing leaders begging her to resign. A day after her boss told her only revenue source to go fuck itself, Yaccarino shared a memo with Twitter employees that she shared in Musk’s “unmatched and completely unvarnished perspective and vision for the future,” claiming that Musk’s “candid and profound” interview was all part of his plan to be an “open platform without censorship of thought,” and that a website that was hemorrhaging revenue and users while becoming an increasingly unreliable source of information was “the most consequential platform that exists.”
Yaccarino, like Sisyphus, must continue to try and push revenue into a site that Musk makes harder to monetize seemingly every day. Her last desperate act is to try and focus Twitter’s sickly advertising business on smaller and medium-sized businesses, like outsourcing sales and “adopting an automated “self-service” small business platform,” which is the kind of thing you’ll hear a desperate startup founder say when you ask them what their plan is to make revenue before they’ve come up with one. Yaccarino is living in Hell, trumpeting that “[X’s] principles do not have a price tag, nor will they be compromised — ever,” though never really specifying what those principles are or to whom they apply to.
You may believe it can’t get worse, and you’d be completely, utterly wrong.
Where’s Your Ed At is a free newsletter, but if you like my work and want to kick me a few dollars, you can do so here.
Like reading Where’s Your Ed At? Perhaps you’d like to join There’s Your Ed At, our Discord Chat Room? You can find it at chat.wheresyoured.at or at this link. It’s free, and a great place to talk with other readers (and me, of course).
Last week, Musk had said that “all” X Premium Plus subscribers would get access to “Grok,” a “rebellious” ChatGPT competitor with “fewer guardrails” that Musk has said was trained on Twitter’s own data, something that Microsoft once tried, creating the world’s most racist chatbot in less than 24 hours back in 2016.
Musk outright lied, saying Grok is “being opened up slowly to Premium+ users,” a statement he likely made because a popular account posted that Grok was a feature of Premium+ subscriptions, only to be met with a community note saying that “most users with X Premium+ still lack access to Grok,” despite Musk posting two days beforehand that you should “subscribe to Premium+ for no ads and access to our Grok AI.”
Regardless of when he does so, Musk’s plan is to launch his anti-woke ChatGPT competitor to Twitter’s user base, train it on Twitter’s data, and then allow it to “analyze” posts, providing some sort of hackneyed (and, based on his example, quite bigoted) rundown of a post or thread, assuming that this isn’t just another one of Musk’s vaporware ideas, like Twitter’s Content Moderation Council, Tesla’s magical, unbreakable brake pads, or, of course, the new Tesla Roadster, which was meant to be available in 2020.
Grok is apparently trained on billions of posts and allegedly has real-time access to X’s data, a form of digital inbreeding that will continually train its model on the data of a website that, other than being a deeply-unreliable source of information, is beset with spam, both from the numerous bots on the site and Twitter Blue users desperate to juice its creator program. While it’s possible that those working at xAi have found ways to mitigate this lack of quality, it is, from what I can tell, the only real-time data that Grok has access to.
In the event that Grok is truly trained on Twitter’s posts (after all, this is an Elon Musk product), it will become what Jathan Sadowski calls a “habsburg AI,” a “system that is so heavily trained on the outputs of other generative AI's that it becomes an inbred mutant, likely with exaggerated, grotesque features.”
Musk rushed a Large Language Model to market in a few months, one where bigotry is a feature not a bug, and has potentially trained it on the garbled musings of people that he has incentivized to fill its data source with nonsense. And that’s acknowledging the fact that LLMs — even those developed a legion of well-resourced and well-fed PHDs, like GPT-4, and based on a well-curated dataset of high-quality material — have a habit of hallucinating facts, figures, and events. Already, despite only being available to a handful of users, Grok has fabricated a story about a British MP being expelled from Parliament, presumably confusing the ouster of disgraced New York congressman George Santos with a separate and unrelated story from across the Atlantic.
One particular problem Grok will have is if it ever publishes anything legally-actionable to X.com in a way that is readily-accessible to a regular user of the site (versus in the form of a screenshot). Doing so will mean that Twitter no longer has protection under Section 230, as said content wouldn’t be considered user-generated, and would therefore be held liable if the subject decided to commence defamation proceedings.
It’s also worth noting that, on a global scale, Section 230 is a bit of an aberration. Most countries don’t offer online platforms such a generous get-out-of-free card as the United States does. Australia — which is the defamation capital of the world — has previously ruled Meta and Alphabet responsible for the activities of their users. In 2022, the latter was ordered to pay $500,000 to a senior New South Wales politician over YouTube videos a judge deemed defamatory, and to which Alphabet was ruled to be a publisher — not a platform. The United Kingdom has similarly-punitive defamation legislation.
This is, of course, an extremely obvious thing to avoid, meaning that Musk is likely to try it based on his outright hostility toward journalists. If he believes that Grok can be a competitor to the legacy media companies he’s claiming are trying to kill his website, he may very well see this as an opportunity to “fight back” and add a section to the website “powered by Grok” that generates publicly-available misinformation at a massive scale.
Even if it doesn’t, it will likely further pollute an already-murky pool of information, the previously-mentioned generative inbreeding churning out useless nonsense that will, in turn, be posted by his gormless fans, which will in turn train Grok further.
And yet the biggest danger that Musk faces with Grok is that Large Language Models are incredibly expensive to build and maintain. Musk apparently purchased 10,000 “data-center grade” GPUs back in April of this year, with said chips costing somewhere in the region of $10,000 a piece, and the program likely being run out of one of Twitter’s two data-center sites in Atlanta and Oregon. ChatGPT allegedly costs around $700,000 a day according to analyst Dylan Patel, and that’s likely with a team of DevOps professionals doing all they can to optimize the performance, versus the skeleton crew left behind to try and make Twitter a usable product.
Editor’s note: A previous version of this article said that ChatGPT cost 36 *cents* a query versus .36 cents - and estimated Grok’s use based on that. It’s fixed now.
Musk is already using Grok as a marketing vehicle for X Premium+’s $16-a-month subscription, and if he actually offers it to every subscriber, there’s no way in hell that this is a profitable transaction. Dylan Patel’s analysis of ChatGPT suggested that it costs OpenAI around .36 cents a query (a cost cushioned by Microsoft’s cloud credits that drove Altman to try and found an AI-focused chip company), based on the cost of running 3600 A100 AI servers, which house around 28,936 GPUs to serve ChatGPT (published in February 2023, around two months before news broke of Musk’s GPU acquisitions, making it likely they use similar hardware). If Grok costs even a quarter of what ChatGPT does a day, that’s still $175,000 a day from a company that said a year ago it was losing $4 million a day, and that’s before they lost such a remarkable amount of advertising revenue.
If Musk launches Grok to every Premium+ subscriber, it creates an inverse revenue model where the site becomes less profitable as its products become more popular. Musk’s insistence on creating and owning a Large Language Model (rather than connecting to another LLM like, say, Snap, a company that has had exactly one profitable quarter) means that he is responsible for driving down costs rather than an external vendor, something that he has proven to be incapable of doing without destroying everything in the process. On top of that, he’s actively marketing the product by asking Grok all manner of stupid questions, meaning that his mewling sycophants will spend days dicking around with it.
As I’ve said before, Musk’s “sell the sizzle” strategy is antithetical to Twitter’s future, because Twitter’s real value isn’t in its product but in hosting a real-time feed of the world’s thoughts. Grok is financial cancer for an already-dying company, and may precipitate the actual death of Twitter — or, as Musk has threatened in the past, its bankruptcy.
Shot By The Albatross
As it stands, the banks behind the Twitter acquisition can’t actually force Elon Musk to sell, and they can’t sell the debt themselves without taking severe losses. Musk currently pays more than a billion dollars a year servicing the debt of a website that was on course to make $3 billion in revenue in 2023 as advertisers had already begun to pull out of the platform. In July Musk would remark that advertising revenue was down 50%, and then rebrand the site to “X,” which would wipe billions from its valuation in the process. While one might argue that attaching Grok to the website might increase its value, it’s unlikely to do so by much. Twitter is only worth $19 billion as of October, and Twitter doesn’t own their Large Language Model — xAI does.
Musk could, of course, merge the companies, but doing so would — much like training Grok on xAI’s data — corrupt xAi in the same way that it has corrupted Musk. Social media companies are deeply fragile businesses, and Musk’s acquisition was a combination of hubris and desperation, a marriage of bigotry and legal missteps that involved little forethought and an almost immediate erasure of anybody who had any real domain expertise in running this kind of company.
The unique circumstances of both the acquisition and how social networks function combined with Musk’s arrogance and Dunning-Kruger lifestyle mean that he is simply incapable of running this company properly. Twitter went private, meaning that there was no way for Musk to arbitrarily pump its value. Social networks are generally not monetizable outside of advertising, which requires executives to have the most bland morals possible, for fear of scaring off their only source of revenue.
Musk clearly has rabid, unquestioning fans, but he has vastly overrestimated how many of them would pay for a website that they’ve predominantly used for free. Just over 800,000 had signed up for X Premium+ as of August 2023, about a month after Musk launched a program where he’d share advertising dollars, another quasi-marketing expense that eats away at Twitter’s already-thin revenue. His only friends are just like him — disloyal right-wing firebrands with specious, paper-thin views borne of a lack of accountability and a theatrical sense of victimhood.
I predict that Musk’s sanity will deteriorate as things sour at Twitter in parallel with the disastrous launch of Tesla’s Cybertruck in 2024. This week saw Tesla reveal its pricing and delivery windows, with the first deliveries of the $70,000 and $97,000 versions (allegedly) starting in 2024 — a 30% increase in price for a truck with 30% less range. Musk claims that two million people have preordered the truck, and as WIRED’s Carlton Reid notes, if Tesla can succeed in selling 50,000 of them it’ll be considered a success. If it doesn’t, it will potentially kill the company, because the Cybertruck won’t be profitable “until 2025” according to Elon Musk, the least-reliable source of information on his own companies. More-insultingly, Tesla will charge you if you resell the truck within the first year.
Another problem, as Andrew J. Hawkins of The Verge added, the world is a little different than it was in 2019, both economically and in the sense that there are now multiple electric trucks on the market at a time when the EV market is beginning to cool, especially for EV trucks.
And when you put aside the fact that this a mistimed launch of a product nobody asked for, you have to also consider that it looks fucking terrible. In early November, Daniel Golson of InsideEV saw the Cybertruck up close, he noted multiple huge panel gaps, misalignments in the fender flares, and that the truck likely would have terrible rear visibility — a big requirement of a vehicle used to tow other vehicles. An image circulated recently of the cybertruck shows that its truck bed is too small to fit a bicycle, and Marques Brownlee revealed that not only does the cybertruck not have any dashboard behind the wheel (requiring you to look at the center screen instead), you can’t see out the rear view mirror if you have the truck bed closed, requiring you to look, again, at the screen. Tesla has clearly made sure to cater to their sycophants with these previews, with Top Gear and Hagerty both effectively operating as marketing extensions for the company while Marques Brownlee constantly contorting to avoid saying anything remotely negative.
It’s also likely quite dangerous, with Kea Wilson of Streetsblog USA calling its hood design “blunt, flat…[and] essentially guaranteed to toss a walker or cyclist under the car’s wheels,” and saying that “the guts of the Cybertruck tick off a few more boxes on the how-to-design-a-car-that-kills-everyone-outside-it checklist.”
The End Of Elon
While there are few things that will actually destroy Elon’s empire, I believe the next year will begin its collapse, if only because he is utterly incapable of admitting he’s wrong.
Twitter has become his own personal 9th Gate. He believed that, despite not having any real domain expertise, he could turn it into a revenue-generating machine, never once considering that his ideas were bad, or that his beliefs were flawed, or that he perhaps wasn’t acting rationally. When things get worse, Elon Musk doesn’t get smarter or better — he crumbles under pressure, becoming churlish and reacting as quickly as possible, unburdened by good counsel or self-awareness to the point that his only solution to a problem is to create a new one.
Twitter as a business and as a website is unique, in that it deeply punishes this kind of attitude. One can improve Tesla’s health by selling more vehicles, or SpaceX’s by raking in government subsidies, but there is no real product for Twitter to sell other than premium versions of a thing that people are used to getting for free. Musk was half right that Twitter’s value lay in its data, but misunderstood that every effort to monetize that data — by selling people the right to prioritize their posts, for example — destroys the egalitarian value of a free-for-all firehose of the world’s thoughts. The only other valuable part was in its advertisers and the reliability of said data, and both have been so thoroughly damaged it’s impossible that Musk can fix it without stepping away from the site for good.
The problem is that he doesn’t believe he’s doing anything wrong. He didn’t delete the antisemitic post that caused advertisers to leave (I’m not linking to it), and he wasn’t capable of apologizing without slinging insults. He blames everybody and anybody else for problems that he himself has caused — “brittle code stack,” “activists,” “bots,” AI scrapers, the ADL, Media Matters and the “woke mind virus” for Twitter’s problems, a private school for turning his trans daughter against him, and the Federal Reserve’s interest rate hikes for Tesla’s drop in market cap. I cannot find a single instance where the richest man in the world has ever actually admitted wrongdoing and did something to make amends. Everything is always somebody else’s fault, resulting in dangerous, destructive tantrums.
He has only a few options with Twitter. He could dig deeper into the dark hole he’s created, catering to the worst people alive, leaving its userbase a combination of fellow right-wing nutjobs, sycophants, and AI spammers, but doing so would require him to continually make billion-dollar payments to increasingly-upset bankers. He could renegotiate the payments, but what would the banks have to gain from him doing so? They’ve already been burned numerous times, and while it’s possible that they might, they could just as easily tell him to pound sand.
In any case, Musk is going to have to continue to pay both the interest on the debt and make up for Twitter’s massive shortfalls in revenue, and may be forced to sell more Tesla stock at a time when the company is beginning to look quite fragile. He could take SpaceX public or spin off Starlink for an IPO (though he’s denied he’d do so), but that would bring more scrutiny to his sole stable-looking company. Much like he ransacked Twitter’s office for stuff to sell, Musk has already made desperate moves like borrowing $1bn from SpaceX, and I predict more is to come.
He could, of course, let the site go bankrupt, but doing so would lead to the kind of discovery, as James Clayton of the BBC noted, that he was trying to avoid by buying the site in the first place. Doing so would trigger a fire sale on Twitter’s debt, and really begin the end of his legacy — but if I’ve learned anything from the last 3 years of history, it’s that stupid shit is always possible.
This is Elon Musk’s death spiral, the beginning of the end of a man so wealthy that he hasn’t experienced a real problem in decades, who has become so neurotic and reactionary that he can only make things worse. His future will be riddled with humiliation and destruction, and I can only hope the collateral damage isn’t severe.