Elon Musk Still Wants Everyone (Including The Judge) To Believe His Fight With Twitter Is About Spam. It’s Not
from the that's-not-how-any-of-this-works dept
As I type this, I’m sitting in a (fairly uncomfortable) chair in the lobby of a Holiday Inn, having read through nearly 300 pages of legal filings of sniping between Elon Musk (165 pages) and Twitter (127 pages) trying to figure out how to best explain what’s in the filings in a meaningful and accurate way. Because the media coverage of this case continues to suck. For example, you may have heard that Elon Musk “countersued” Twitter. Headlines blasted that left and right and Musk’s fans lapped it up. I saw multiple tweets claiming that Musk was going to cost Twitter “so much” money by suing them back.
The reality is… less interesting than that. Musk filed his required answer and defenses to Twitter’s complaint, and, with it, included a bunch of counterclaims. This isn’t particularly surprising. It happens pretty frequently, especially in business disputes. While there are technically claims that Twitter has to respond to now, it’s all part of the same lawsuit. Also, the claims are ridiculous and clearly involved Musk’s lawyers scraping whatever dignity they had off the floor to try to come up with something… anything they could include as counterclaims.
There was a fun little side story to all of this in that Musk filed the answer and counterclaims under seal, as was required, giving Twitter some period of time to review the document to see if it included any confidential information that it wanted redacted before the document was made public. Apparently, Musk sought to speed up the date by which a public version would be released, and Twitter demanded the full five business days that it was entitled to, in order to review the document to highlight what needed to be redacted. The judge granted Twitter’s request, meaning that even though the document was filed under seal on July 29th, it wasn’t actually made public until August 5th.
But… and this is kinda funny, actually: Twitter was able to file its unredacted reply to Musk’s filing before his redacted filing was made public. And then after all that Twitter said it didn’t need any of Musk’s filing to be redacted in the first place. It’s almost as if Twitter was able to use this process to make sure that its responses to Musk’s claims were revealed effectively simultaneously with Musk’s claims. This feels a bit like playing games with the system — something courts sometimes don’t take kindly to — but given all the games Musk appears to be playing, this seems somewhat justified.
Within all of these pages there are many, many stories worth telling, and indeed, when I’m not in a Holiday Inn lobby, I’ll probably write a few more posts about certain aspects of these filings, but for now I have just a few thoughts. Again, these will not cover everything in the filings, at all. But just a few of the key things.
Musk really wants everyone, including the judge, to believe this is about spam on the platform. It’s not. Honestly, it feels like Musk’s best bet for succeeding in this case is for the judge to be an idiot, and be taken in by Musk’s nonsensical claims like many of his loyal fans. This feels like a long shot, but he’s had so much success with his cult of personality nonsense, that maybe he can pull it off one more time. I have gone into great detail as to why this case has nothing whatsoever to do with spam, so I’m not going to do it all over again.
Suffice it to say, Musk’s officially stated reason for breaking the deal is the (completely unsupported) claim that Twitter breached the purchase agreement by not giving him the information he needed to complete the deal. That’s the dispute. Musk has been trying to turn this into a story about spam, because that’s the best excuse his very expensive lawyers could come up with to try to get him out of this deal. Musk signed the deal and waived due diligence rights. He also could have had Twitter put in representations regarding the amount of spam on the platform if he was really concerned by it.
Hilariously, in his filing, Musk points to a text message conversation he had with Twitter CEO Parag Agarwal highlighting his concern about spam on the platform.
Musk uses this screenshot as part of his argument for why he wanted to buy Twitter… but it seems to undermine the very core of his own argument. Note that the date of the conversation is April 8th. That’s before he bought Twitter. Hell, that’s actually back when it was still expected that Musk would be joining Twitter’s board. It was a day before he rejected that plan, and announced his intention to purchase the site outright, a proposal that ended up with a signed agreement a little over two weeks later.
So… that sure seems like proof that Musk was concerned about spam on the platform — and if that’s the case then (1) it sure seems like he considered himself well aware of the amount of spam on the site before the deal was consummated, and actually argues in this very document that he thought he could unlock more value for the company by getting rid of the spam. If that were true, having more spam on the platform than previously believed wouldn’t much matter. He could just get rid of it all and do all the other improvements he wanted’ More importantly (2) if this were such a big concern he could have easily demanded information about it prior to doing the purchase and required that Twitter make representations regarding the amount of spam on the site which, if they turned out to be false, could scuttle the deal.
He did none of that.
Instead, he makes a fantastical case, setting out all the other stuff he’s been saying publicly about spam, to argue that Twitter committed “fraud.” But he fails to actually make a case for that. It’s all random speculation that doesn’t actually add up to anything.
He claims, repeatedly, that he relied on Twitter’s SEC filings about how much spam was on the platform, but the message to Parag undermines that. What undermines it even more is that just weeks after the deal was signed, Musk appeared at a conference and said “anyone who uses Twitter is well aware that the comment threads are filled with scam, spam, and a lot of fake accounts. It seems beyond reasonable for Twitter to claim that the number of real unique humans is above 95%.” In other words, here is Musk, soon after the deal was announced, admitting that “anyone who uses Twitter” must be aware that the amount of spam on the site is greater than what Twitter reported to the SEC.
In short, here he is admitting that at least he long believed the amount of spam on the site was way above what the company reported to the SEC, and therefore, he has admitted that he certainly didn’t actually rely on Twitter’s filings with the SEC (which, again, do not say that spam itself is less than 5% of the service — just that less than 5% is counted in the mDAU number the company reports).
Musk’s lack of understanding of some fairly basic concepts is embarrassing. Musk is supposed to be this ubergenius, right? I mean, even as I’ve sometimes criticized some of the things he’s said over the years, I definitely believed he was still incredibly smart and innovative in his thinking. But the more I see about how this whole Twitter scenario has played out, the more I wonder how he was able to succeed at anything. Over and over again in this filing, Musk basically reveals his ignorance of how some fairly basic things work.
We’ve already discussed, multiple times, in the past why Musk’s freak-out over Twitter having human reviewers look at 100 accounts per day, and not using “AI” is completely confused and misplaced. You can read that again to understand why Twitter’s statistical sampling is actually quite good here. But here’s the short version again: Musk keeps complaining that Twitter doesn’t use automation and AI to determine how much of its monetizable daily active users (mDAU) were actually spam accounts. But… that completely ignores reality, which Twitter explained both directly to Musk and to the public.
Twitter DOES use automation and AI to ban or suspend potential spam accounts. It does that on hundreds of thousands of accounts EVERY DAY. It’s only after that, that Twitter then does a spot check, by having human reviewers look at a random sample of 100 accounts per day. And while it may sound like 100 accounts per day seems like a small number, that’s only if you don’t understand statistics. 100 accounts per day is 9000 accounts per quarter (the period reviewed) and that can give you a 1% margin of error with 95% confidence. That’s good.
But, more importantly, if Twitter used AI and automation to check what was spam… the answer would be zero. Because — as Twitter and others have explained multiple times — this human review takes place after the AI and automation have already removed all the spam it could find. And that’s basically the only way you could do this. If you have automation and AI finding spam, then you shut down those accounts. The whole point of human review is to see what’s left afterwards. And you can only do that with human review, because otherwise, why wouldn’t you have used that AI and automation to find the spam in the first place.
Musk’s entire argument — repeated throughout the filings makes no sense. Twitter does use AI and automation to discover spam. It’s been quite public about that. It then checks with humans using a statistically significant sampling to determine how much spam is still there. That’s it. But Musk and his lawyers seem mentally incapable of understanding this seemingly simple fact.
Musk’s own method for calculating spam accounts is laughably bad. Musk complains a lot about how much spam he thinks is on the platform, but provides scant evidence to support his claims that Twitter’s numbers are false (and again, it doesn’t matter if they’re false, because the agreement doesn’t turn on that, but even so, he’s got no evidence). While he complains throughout the filing that Twitter didn’t give him the information he needed to replicate Twitter’s system for judging spam, he’s playing some tricks here.
The reviews are human reviews, and so Musk is basically asking for a download of Twitter employees’ brains explaining how they went through the 100 accounts that day and determined how many were spam. That’s not data you can just hand over (also, arguably, it’s also data Musk has no right to at this point under the agreement — as the information sharing is only supposed to be for the purpose of closing the deal, and Musk makes no argument as to how this is actually necessary to close the deal — perhaps because it’s not).
While Twitter does hand over a lot of information to Musk, the only attempt by Musk to prove the higher spam numbers is… to use Botometer.
The Musk Parties’ analysis processed accounts visible on the Firehose using the University of Indiana Botometer tool, which was initially developed with support from the DARPA program and has been improved and honed over the past eight years. The academic developers of the Botometer tool have published numerous articles about their work, including one seminal paper that has received over 1,000 citations in the academic literature. Defendants’ experts are continuing their analysis even now and, in anticipation of production of additional data by Twitter (including “private” data that Twitter makes available to its human reviewers and contends is necessary to verify its reported less-than5% spam and false user rate), intend to conduct a more comprehensive analysis and expect to present updated estimates and findings in expert reports and at trial.
Um. WTF? Really? There are a few services out there that claim they can tell if accounts are bots… and they’re fun to play with, but they’re all toys. They’re not accurate at all. And this point is really emphasized in Twitter’s reply when it notes that MUSK’S OWN ACCOUNT was recently deemed by Botometer to be a bot.
Musk’s “preliminary expert estimates” are nothing more than the output of running the wrong data through a generic web tool. The data are wrong because, as Musk knows, the “Firehose” from which the data were collected reflects many Twitter accounts that are not included in mDAU and, at the same time and as Musk admits in footnote 2, does not reflect the majority of those accounts that are included in mDAU because those accounts, though logged in, are not Tweeting or taking other actions. Confirming the unreliability of Musk’s conclusion, he relies on an internet application called the “Botometer”—which applies different standards than Twitter does and which earlier this year designated Musk himself as highly likely to be a bot.
Later, Twitter’s lawyers go further:
Insofar as Defendants rely on the Botometer for their analysis of spam, Twitter avers that the Botometer’s own FAQ website cautions that “Bot detection is a hard task” and that if it “were easy to do with software, there wouldn’t be any bots—Twitter would have already caught and banned them!” Twitter further avers that in May 2022, protocol.com reported that the Botometer indicated that Elon Musk’s own Twitter account was likely a bot, scoring it 4/5.
If you want to have some fun look over the discussion on Twitter between the guy who noticed that Botometer called Musk a bot and Botometer, in which it explains why it’s so difficult to algorithmically determine botness.
But, yeah, this image from Duke Professor Chris Bail is just… perfect:
Musk also doesn’t seem to understand some fairly basic things about internet business models. Okay, look, so Musk’s big success stories haven’t really been about the internet. Yes, you can argue that PayPal was, but it was more financial services, and Musk was apparently long gone from the company before it was actually a big success.
But, a big part of Musk’s filing focuses on the fact that there are a small number of super valuable Twitter accounts that make Twitter a lot of money. Then there is a decent chunk that makes Twitter a decent amount of money — and there are a bunch that don’t really make much money at all.
But, um, this is just how internet business models work. No one with any knowledge of any of this thinks that all users are equally valuable. Of course, you have some users who are much more valuable than others. That’s just common sense.
Yet there are pages upon pages of Musk and his lawyers insisting that this basic level of knowledge has been somehow hidden by Twitter, and that’s evidence of fraud or some such. It’s… just weird.
Musk’s own earlier filings undermine his latest counterclaims. Throughout the filing, beyond Musk’s (and his lawyers’) fundamental misunderstanding of what Twitter has reported to the SEC and how that impacts everything else, Musk keeps claiming that Twitter simply refused to explain to Musk the process that the company used to do that human audit of 9,000 random accounts every quarter. Over and over again the filing says that Twitter “refused” to provide the details of how those audits worked.
Indeed, over the weekend, Musk claimed that “if Twitter simply provides their method of sampling 100 accounts and how they’re confirmed to be real, the deal should proceed on original terms.”
Except, as Chancery Daily has pointed out, in Musk’s own filings, he filed a sealed exhibit, entitled “mDAU Auditing Guidelines (Confidential)” showing that Twitter already provided him with exactly what he’s claiming they never gave him. And he gave it to the court:
This is actually even worse than it seems. This sealed exhibit was included with Musk’s attempt to push back on Twitter’s motion to expedite the trial (which didn’t work). In that filing, the one citation to Exhibit 4 is… where Musk claims he was “stunned to discover” that Twitter’s process for spot checking how many spam accounts get through its automation didn’t use automation (again, he falsely characterizes this as how much spam is on the platform, even though, as explained above and elsewhere repeatedly, it’s not that).
But, while the full document remains sealed, it’s now clear that Twitter did, in fact, give Musk their guidelines for mDAU auditing. He may not like those guidelines. But in his counterclaims he argues that Twitter didn’t provide them. Yet, clearly the company did. And Musk already knows that, because he gave the court the document!
Anyway, the big takeaways from this are that Musk is still playing a game in which he’s trying to baffle the judge with bullshit. It may work for some of his fans, but the only person it actually matters for here is the judge: Chancellor Kathaleen McCormick. Chancellor McCormick has a reputation as a no-nonsense judge, and it would be surprising to see her taken in by all this.
Anyway, that’s all for now. As I noted, there are some other interesting things in the filings, but I think they require separate posts which I’ll hopefully write up eventually from a more comfortable chair.