Why is Musk mad about Community Notes?
🤥 Faked Up #38: Musk says Community Notes are "being gamed"; AI slop is the lure for a scam operation; and TikTok videos kill Pope Francis.
If you're a free subscriber, you did not receive the preview of last week's newsletter because I clicked the wrong button. Apologies! Catch up here.
Paid subscribers: I'm trying a new format for the final links section. Could you tell me whether you like it or prefer the way it was before (e.g. FU#36)?
This newsletter is a ~7 minute read and includes 72 links.
NEWS
President Milei of Argentina faces fraud charges over a crypto tweet. 404 Media shamed an ebook service into purging its AI slop. Meta, TikTok, and X testified about misinformation to the UK Parliament. The head of South Korea's Constitutional Court is being targeted by disinformation as he deliberates over the impeachment of president Yoon Suk Yeol. Rumble has been suspended by a Brazilian judge for not acting on misinformation days after suing the judge in Florida alongside Trump Media (it's complicated). Incoming deputy FBI director Dan Bongino was an InfoWars regular. A preprint about Covid-19 vaccine side effects was received as calmly as you'd expect. Serbian police raided the offices of a local fact-checker, allegedly to investigate USAID-related corruption. Google Search quietly killed data void warnings (💔).
TOP STORIES
Elon Musk slams Community Notes and muzzles Grok
On Thursday, Elon Musk said that X's crowdsourced fact-checking intervention Community Notes "is increasingly being gamed by governments & legacy media" and that he is "working to fix this…"
This may have come as a surprise to the engineering lead for Community Notes, who around the same time posted that the product was "pretty robust against coordination attempts."
Musk has been upset with Community Notes for a few days, perhaps because he discovered that they extensively cite Wikipedia and fact-checking websites. He has taken to posting screenshots of tweets so as to feature Community Notes that have not been deemed helpful (but presumably align with his views).
The precipitating factor for his accusation that the feature is "being gamed" appears to have been notes about the legitimacy of Ukrainian president Volodymyr Zelensky. Thirty minutes before Musk's tweet on the topic, Donald Trump had posted about it. At the time of writing, that tweet had more than forty proposed notes, but none have been deemed "helpful" by the Community Notes algorithm.
Notes deemed "helpful" by the right mix of users did appear on prior tweets criticizing Zelensky by Donald Trump Jr. (who reshared an altered video) and Republican Senator Josh Hawley (for a claim about auditors).* It's possible that Musk posted about manipulation because he saw these notes.
Incidentally, the notes on Trump Jr.'s tweet are no longer visible. I turned to my Community Notes go-to expert Alex Mahadevan to help me figure out why.
Alex says all three notes were public at one time but reverted to "needs more ratings" on Friday. The top notes was a straightforward debunk with a link to a fact check (see below). It was rated "helpful" by 231 users and "not helpful" by only 15.
Why is it no longer visible? Thanks to the admirable transparency underlying the program, we can answer that question. X's topic model, which requires people who disagree about the specific topic of the tweet to agree about a note's helpfulness, flipped the note to "needs more ratings" at some point on Thursday afternoon.
The topic model was a more stringent consensus requirement introduced after Musk took over at X. If its impact on this tweet and Musk's displeasure with notes is anything to go by, we may start seeing even fewer notes on tweets than the ~10% that currently please the consensus algorithms. This is the best case scenario; it is not impossible to envision Musk asking his engineers to reverse engineer an algorithm that features "governments & legacy media" less.
Some engineers may comply in advance. Last week, xAI staff briefly censored Grok after it called out Musk for spreading disinformation. According to xAI's head of engineering, this was down to an "employee pushing a change to the prompt that they thought would help without asking anyone at the company for confirmation."
The employee had hard-wired an instruction for X's AI text generator to "ignore all sources that mention Elon Musk/Donald Trump spread misinformation."
All this is a reminder that Muskworld's commitment to fighting misinformation is nakedly self-serving. I wonder how all this landed with Mark Zuckerberg, whose grotesque pivot to Community Notes was predicated on pleasing Musk and Trump. Just last week, Meta opened the waitlist for users who want to join its own notes program (I applied). It doesn't take a rocket scientist to predict that if it upsets anyone in power, it will be accused of being partisan.
AI motivational slop is the top of the scam funnel
Two weeks ago, I wrote about AI-generated motivational videos of Elon Musk and Donald Trump racking up 700 million views on TikTok. At the time, I wasn't sure what the content was for. Now, I have an answer: Scams.
Last week, at least five of the accounts I track started posting videos of AI Donald Trump promising viewers millions on a specific date. AI Trump calls on users not to scroll past the video, but to "tap the five buttons" and leave a specific comment that is typically "I deserve it" or "Yes".
The digitally naive folks who did as they were instructed got a reply by one or more TikTok users encouraging them to send a direct message or reach out via WhatsApp "to claim $10,000." They basically self-selected into a scam sting.
In the interest of science, I texted the number in the comments (see below). It had a WhatsApp business account with a green dollar sign as a profile picture, so I knew I was in the right place. My interlocutor asked me for a $100 "activation fee" via Venmo to guarantee my $10,000 reward on CashApp. Once I went quiet, they told me not to get scared because "the victory you see for is inside the battle you fear."
I opted instead to report the account to Meta, which terminated it. I also reported the AI slop channels to TikTok, which deleted four of the five (I have since found at least two other live accounts that are part of the same operation).
TikTok creators killed Pope Francis for the likes
Pope Francis has been hospitalized for the past couple of weeks and appears to be in critical condition. But he is still alive. That hasn't stopped many TikTok creators from claiming that he died, often supported by preposterous AI-generated videos. One video with 9.8 million views called on users to share it before it got deleted. A sizable minority of its almost 11,000 comments was from people who appeared to believe that Francis had actually passed.
(TikTok deleted all the misleading videos that I flagged to it on Tuesday.)
Inevitably, there was also a "Rest in peace pope" video showing Francis getting a hug from AI Jesus: