Latest news with #JoelKaplan
WIRED
29-05-2025
- Business
- WIRED
Meta's ‘Free Expression' Push Results In Far Fewer Content Takedowns
May 29, 2025 7:09 PM Meta says loosening its enforcement policies earlier this year led to fewer erroneous takedowns on Facebook and Instagram—and didn't broadly expose users to more harmful content. An aerial view of Meta headquarters in Menlo Park, California. Photograph:Meta announced in January it would end some content moderation efforts, loosen its rules, and put more emphasis on supporting 'free expression.' The shifts resulted in fewer posts being removed from Facebook and Instagram, the company disclosed Thursday in its quarterly Community Standards Enforcement Report. Meta said that its new policies had helped reduce erroneous content removals in the US by half without broadly exposing users to more offensive content than before the changes. The new report, which was referenced in an update to a January blog post by Meta global affairs chief Joel Kaplan, shows that Meta removed nearly one third less content on Facebook and Instagram globally for violating its rules from January to March of this year than it did in the previous quarter, or about 1.6 billion items compared to just under 2.4 billion, according to an analysis by WIRED. In the past several quarters, the tech giant's total quarterly removals had previously risen or stayed flat. Across Instagram and Facebook, Meta reported removing about 50 percent fewer posts for violating its spam rules, nearly 36 percent for child endangerment, and almost 29 percent for hateful conduct. Removals increased in only one major rules category—suicide and self-harm content—out of the 11 Meta lists. The amount of content Meta removes fluctuates regularly from quarter to quarter, and a number of factors could have contributed to the dip in takedowns. But the company itself acknowledged that 'changes made to reduce enforcement mistakes' was one reason for the large drop. 'Across a range of policy areas we saw a decrease in the amount of content actioned and a decrease in the percent of content we took action on before a user reported it,' the company wrote. 'This was in part because of the changes we made to ensure we are making fewer mistakes. We also saw a corresponding decrease in the amount of content appealed and eventually restored.' Meta relaxed some of its content rules at the start of the year that CEO Mark Zuckerberg described as 'just out of touch with mainstream discourse.' The changes allowed Instagram and Facebook users to employ some language that human rights activists view as hateful toward immigrants or individuals that identify as transgender. For example, Meta now permits 'allegations of mental illness or abnormality when based on gender or sexual orientation.' As part of the sweeping changes, which were announced just as Donald Trump was set to begin his second term as US president, Meta also stopped relying as much on automated tools to identify and remove posts suspected of less severe violations of its rules because it said they had high error rates, prompting frustration from users. During the first quarter of this year, Meta's automated systems accounted for 97.4 percent of content removed from Instagram under the company's hate speech policies, down by just one percentage point from the end of last year. (User reports to Meta triggered the remaining percentage.) But automated removals for bullying and harassment on Facebook dropped nearly 12 percentage points. In some categories, such as nudity, Meta's systems were slightly more proactive compared to the previous quarter. Users can appeal content takedowns, and Meta sometimes restores posts that it determines have been wrongfully removed. In the update to Kaplan's blog post, Meta highlighted the large decrease in erroneous takedowns. 'This improvement follows the commitment we made in January to change our focus to proactively enforcing high-severity violations and enhancing our accuracy through system audits and additional signals,' the company wrote. Some Meta employees told WIRED in January that they were concerned the policy changes could lead to a dangerous free-for-all on Facebook and Instagram, turning the platforms into increasingly inhospitable places for users to converse and spend time. But according to its own sampling, Meta estimates that users were exposed to about one to two pieces of hateful content on average for every 10,000 posts viewed in the first quarter, down from about two to three at the end of last year. And Meta's platforms have continued growing—about 3.43 billion people in March used at least one of its apps, which include WhatsApp and Messenger, up from 3.35 billion in December.
Yahoo
10-05-2025
- Business
- Yahoo
Robby Starbuck calls Meta apology ‘sort of bizarre' amid defamation lawsuit
Robby Starbuck, a conservative author and activist, took a swing at Meta on Friday, just a week after an executive apologized for the company's artificial intelligence (AI) search engine spreading false information about him. 'I think it's one of the first times I've ever seen a company come out in the middle of active litigation and essentially say, 'Yeah, we did that. Sorry about that' — which is sort of bizarre, leaves one scratching your head, like that's a bizarre legal strategy,' Starbuck told NewsNation's 'On Balance' in an interview. 'But essentially, they admitted guilt,' he continued. 'So, in one hand, I'm happy they did that, but secondarily … I'm concerned about what is the precedent for the future. Are we going to allow AI to invent whatever it wants out there as the truth, when, in fact, it's not?' The conservative film maker has accused Meta's system of falsely reporting that he was present at the Jan. 6, 2021, Capitol riot and that he pleaded guilty to disorderly conduct. He later filed a lawsuit against the company that owns Facebook and Instagram. Following roughly nine months of litigation, an executive reached out to Starbuck to issue an apology. 'Robby – I watched your video – this is unacceptable,' Joel Kaplan, Meta's chief global affairs officer, wrote in a post last week on social platform X. 'This is clearly not how our AI should operate. We're sorry for the results it shared about you and that the fix we put in place didn't address the underlying problem.' Starbuck said while he appreciates the apology, he is focused on the bigger picture, including how AI and misinformation could affect elections in the future. 'It comes after us privately trying to solve this, not just for me, but for everybody, because that's my real goal here — I want to fix this problem, so it never does this to anybody,' he told host Leland Vittert, while also questioning why there isn't more bipartisan support for his cause. 'I want to stop this for everyone, regardless of their party,' Starbuck added. Asked if he's seeking damages, the conservative replied 'yeah.' 'There's damages for sure. I mean, it's like any other defamation case,' he said. 'They damaged my reputation, my character.' 'People have come up on the streets thinking that I was a criminal that pled guilty to a crime that I obviously did not commit,' Starbuck continued. 'It even suggested my kids be taken away from me by authorities. And again, my kids have seen people come up to me thinking that I was at Jan. 6 … that I got arrested for breaking in … none of that is true.' He's also reported death threats associated with the false information. According to the lawsuit, Starbuck is seeking over $5,000,000 in compensatory and punitive damages. The legal complaint came after Meta launched its standalone version of its AI assistant late last month, touting more 'personal' and 'relevant' responses. The application uses its Llama 4 coding model to collect information from across the company's platforms, according to the press release. Copyright 2025 Nexstar Media, Inc. All rights reserved. This material may not be published, broadcast, rewritten, or redistributed.
Euronews
08-05-2025
- Business
- Euronews
US companies still engaging with AI Code despite Trump: EU official
US companies do not seem to have changed their attitude towards the EU's Code of Practice on General Purpose AI – which should help providers of AI models comply with the EU's AI Act – after the change in US administration, an official working in the European Commission's AI Office told Euronews. The drafting process of the voluntary Code is still ongoing after the Commission appointed thirteen experts last September, using plenary sessions and workshops to allow some 1,000 participants to share feedback. The final version was planned for 2 May but has not yet been published. The official said the deadline was not met because the Commission 'received a number of requests to leave the consultations open longer than originally planned.' The process has been criticised throughout, by Big Tech companies, but also publishers and rightsholders whose main concern is that the rules are a violation of the EU's Copyright laws. The US government's Mission to the EU sent a letter to the EU executive pushing back against the Code in April. The administration led by Republican President Donald Trump has been critical of the EU's digital rules, claiming that it stifles innovation. Meta's global policy chief, Joel Kaplan, said in February that it will not sign the code because it had issues with the then latest version. The document has been updated since. The EU official said that US companies 'are very proactive' and that there is not the sense that 'they are pulling back because of a change in the administration.' 'I cannot say that the attitude of the companies engaging the code of practice has changed because of the change in the administration. I did not perceive that in the process.' The plan is still to get the rules out before 2 August, when the rules on GP AI tools - such as large language models like ChatGPT - enter into force. The AI Act itself - which regulates AI tools according to the risk they pose to society - entered into force in August last year. Its provisions apply gradually, before the Act will be fully applicable in 2027. The Commission will assess companies' intentions to sign the code, and carry out an adequacy assessment with the member states. The EU executive wants to know 'by August' whether the institutions also accept the code. The EU executive can then decide to formalise the Code through an implementing act. A report by Corporate Europe Observatory (CEO) and LobbyControl published last week suggests that Big Tech companies put pressure on the Commission to water down the Code of Practice, and that they 'enjoyed structural advantages' in the drafting process. In a reaction, the Commission said all participants "had the same opportunity to engage in the process through the same channels." The EU official could not say whether it seems likely that companies will sign, but stressed that it's important that they do. An alternative option, where businesses commit only to specific parts of the Code, has not been put on the table yet. If that becomes a reality, however, they would need to 'fulfil their obligation in a different way,' the official said. The last global gathering on artificial intelligence (AI) at the Paris AI Action Summit in February saw countries divided, notably after the US and UK refused to sign a joint declaration for AI that is "open, inclusive, transparent, ethical, safe, secure, and trustworthy". AI experts at the time criticised the declaration for not going far enough and being "devoid of any meaning," the reason countries cited for not signing the pact, as opposed to their being against AI safety. The next global AI summit will be held in India next year, but rather than wait until then, Singapore's government held a conference called the International Scientific Exchange on AI Safety on April 26. "Paris [AI Summit] left a misguided impression that people don't agree about AI safety," said Max Tegmark, MIT professor and contributor to the Singapore report. "The Singapore government was clever to say yes, there is an agreement,' he told Euronews Next. Representatives from leading AI companies, such as OpenAI, Meta, Google DeepMind, and Anthropic, as well as leaders from 11 countries, including the US, China, and the EU, attended. The result of the conference was published in a paper released on Thursday called 'The Singapore Consensus on Global AI Safety Research Priorities'. The document lists research proposals to ensure that AI does not become dangerous to humanity. It identifies three aspects to promote a safe AI: assessing, developing trustworthiness, and controlling AI systems, which include large language models (LLMs), multimodal models that can work with multiple types of data, often including text, images, video, and lastly, AI agents. The main research that the document argues should be assessed is the development of risk thresholds to determine when intervention is needed, techniques for studying current impacts and forecasting future implications, and methods for rigorous testing and evaluation of AI systems. Some of the key areas of research listed include improving the validity and precision of AI model assessments and finding methods for testing dangerous behaviours, which include scenarios where AI operates outside human control. The paper calls for a definition of boundaries between acceptable and unacceptable behaviours. It also says that when building AI systems, they should be developed with truthful and honest systems and datasets. And once built, these AI systems should be checked to ensure they meet agreed safety standards, such as tests against jailbreaking. The final area the paper advocates for is the control and societal resilience of AI systems. This includes monitoring, kill switches, and non-agentic AI serving as guardrails for agentic systems. It also calls for human-centric oversight frameworks. As for societal resilience, the paper said that infrastructure against AI-enabled disruptions should be strengthened, and it argued that coordination mechanisms for incident responses should be developed. The release of the report comes as the geopolitical race for AI intensifies and AI companies thrash out their latest models to beat their competition. However, Xue Lan, Dean of Tsinghua University, who attended the conference, said: "In an era of geopolitical fragmentation, this comprehensive synthesis of cutting-edge research on AI safety is a promising sign that the global community is coming together with a shared commitment to shaping a safer AI future". Tegmark added that there is a consensus for AI safety between governments and tech firms, as it is in everyone's interest. "OpenAI, Antropic, and all these companies sent people to the Singapore conference; they want to share their safety concerns, and they don't have to share their secret sauce," he said. "Rival governments also don't want nuclear blow-ups in opposing countries, it's not in their interest," he added. Tegmark hopes that before the next AI summit in India, governments will treat AI like any other powerful tech industry, such as biotech, whereby there are safety standards in each country and new drugs are required to pass certain trials. "I'm feeling much more optimistic about the next summit now than after Paris," Tegmark said.
Yahoo
05-05-2025
- Business
- Yahoo
We're Back to the Actually Internet
One Friday in April, Meta's chief global affairs officer, Joel Kaplan, announced that the process of removing fact-checking from Facebook, Threads, and Instagram was nearly complete. By the following Monday, there would be 'no new fact checks and no fact checkers' working across these platforms, which are used by billions of people globally—no professionals marking disinformation about vaccines or stolen elections. Elon Musk, owner of X—a rival platform with an infamously permissive approach to content moderation—replied to Kaplan, writing, 'Cool.' Meta, then just called Facebook, began its fact-checking program in December 2016, after President Donald Trump was first elected and the social network was criticized for allowing the rampant spread of fake news. The company will still take action against many kinds of problematic content—threats of violence, for example. But it has left the job of patrolling many kinds of misinformation to users themselves. Now, if users are so compelled, they can turn to a Community Notes program, which allows regular people to officially contradict one another's posts with clarifying or corrective supplementary text. A Facebook post stating that the sun has changed color might receive a useful correction, but only if someone decided to write one and submit it for consideration. Almost anyone can sign up for the program (Meta says users must be over 18 and have accounts 'in good standing'), making it, in theory, an egalitarian approach to content moderation. Meta CEO Mark Zuckerberg has called the pivot on misinformation a return to the company's 'roots,' with Facebook and Instagram as sites of 'free expression.' He announced the decision to adopt Community Notes back in January, and explicitly framed the move as a response to the 2024 elections, which he described as a 'cultural tipping point towards once again prioritizing speech.' Less explicitly, Meta's shift to Community Notes is a response to years of being criticized from both sides of the aisle over the company's approach to misinformation. Near the end of his last term, Trump targeted Facebook and other online platforms with an executive order accusing them of 'selective censorship that is harming our national discourse,' and during the Biden administration, Zuckerberg said he was pressured to take down more posts about COVID than he wanted to. Meta's abandonment of traditional fact-checking may be cynical, but misinformation is also an intractable problem. Fact-checking assumes that if you can get a trustworthy source to provide better information, you can save people from believing false claims. But people have different ideas of what makes a trustworthy source, and there are times when people want to believe wrong things. How can you stop them? And, the second question that platforms are now asking themselves: How hard should you try? Community Notes programs—originally invented in 2021 by a team at X, back when it was still called Twitter—are a somewhat perplexing attempt at solving the problem. It seems to rely on a quaint, naive idea of how people behave online: Let's just talk it out! Reasonable debate will prevail! But, to the credit of social-media platforms, the approach is not as starry-eyed as it seems. The chief innovation of Community Notes is that the annotations are generated by consensus among people who might otherwise see things differently. Not every note that is written actually appears under a given post; instead, they are assessed using 'bridging' algorithms, which are meant to 'bridge' divides by accounting for what's called 'diverse positive feedback.' This means that a potential note is valued more highly and is more likely to appear on a post if it is rated 'helpful' by a wide array of people who have demonstrated different biases at other times. The basics of this system have quickly become a new industry standard. Shortly after Meta's announcement about the end of fact-checking, TikTok said that it would be testing its own version of Community Notes, called Footnotes—though unlike Meta and X, TikTok will keep using a formal fact-checking program as well. These tools are 'a good idea and do more good than harm,' Paul Friedl, a researcher at Humboldt University, in Berlin, told me. Friedl co-authored a 2024 paper on decentralized content moderation for Internet Policy Review, which discussed X's Community Notes among other examples, including Reddit's forums and old Usenet messaging threads. A major benefit he and his co-author cited was that these programs may help create a 'culture of responsibility' by encouraging communities 'to reflect, debate, and agree' on the purpose of whatever online space they're using. Platforms certainly have good reasons to embrace the model. The first, according to Friedl, is the cost. Rather than employing fact-checkers around the world, these programs require only a simple algorithm. Users do the work for free. The second is that people like them—they often find the context added to posts by fellow users to be helpful and interesting. The third is politics. For the past decade, platforms—and Meta in particular—have been highly reactive to political events, moving from crisis to crisis and angering critics in the process. When Facebook first started flagging fake news, it was perceived as too little, too late by Democrats and reckless censorship by Republicans. It significantly expanded its fact-checking program in 2020 to deal with rampant misinformation (often spread by Trump) about the coronavirus pandemic and that year's election. From March 1, 2020, to Election Day that year, according to Facebook's self-reporting, the company displayed fact-checking labels on more than 180 million pieces of content. Again, this was perceived as both too much and not enough. With a notes-based system, platforms can sidestep the hassle of public scrutiny over what is or isn't fact-checked and why and cleanly remove themselves from drama. They avoid making contentious decisions, Friedl said, which helps in an effort 'not to lose cultural capital with any user bases.' John Stoll, the recently hired head of news at X, told me something similar about Community Notes. The tool is the 'best solution' to misinformation, he said, because it takes 'a sledgehammer to a black box.' X's program allows users to download all notes and their voting history in enormous spreadsheets. By making moderation visible and collaborative, instead of secretive and unaccountable, he argued, X has discovered how to do things in 'the most equitable, fair, and most pro-free-speech way.' ('Free speech' on X, it should be noted, has also meant platforming white supremacists and other hateful users who were previously banned under Twitter's old rules.) [Read: X is a white-supremacist site] People across the political spectrum do seem to trust notes more than they do standard misinformation flags. That may be because notes feel more organic and tend to be more detailed. In the 2024 paper, Friedl and his co-author wrote that Community Notes give responsibilities 'to those most intimately aware of the intricacies of specific online communities.' Those people may also be able to work faster than traditional fact-checkers—X claims that notes usually appear in a matter of hours, while a complicated independent fact-check can take days. Yet all of these advantages have their limits. Community Notes is really best suited to nitpicking individual instances of people lying or just being wrong. It cannot counter sophisticated, large-scale disinformation campaigns or penalize repeated bad actors (as the old fact-checking regime did). When Twitter's early version of Community Notes, then called Birdwatch, debuted, the details of the mechanism were made public in a paper that acknowledged another important limitation: The algorithm 'needs some cross-partisan agreement to function,' which may, at times, be impossible to find. If there is no consensus, there are no notes. Musk himself has provided a good case study for this issue. A few Community Notes have vanished from Musk's posts. It's possible that he had them removed—at times, he has seemed to resent the power that X has given its users through the program, suggesting that the system is 'being gamed' and chiding users for citing 'legacy media'—but the disappearances could instead be an algorithmic issue. An influx of either Elon haters or Elon fans could ruin the consensus and the notes' helpfulness ratings, leading them to disappear. (When I asked about this problem, Stoll told me, 'We're, as a company, 100 percent committed to and in love with Community Notes,' but he did not comment on what had happened to the notes removed from Musk's posts.) The early Birdwatch paper also noted that the system might get really, really good at moderating 'trivial topics.' That is the tool's core weakness and its core strength. Notes, because they are written and voted on by people with numerous niche interests and fixations, can appear on anything. While you'll see them on something classically wrong and dangerous, such as conspiracy theories about Barack Obama's birth certificate, you'll also see them on things that are ridiculous and harmless, such as a cute video of a hedgehog. (The caption for a hedgehog video I saw last week suggested that a stumbling hedgehog was being 'helped' across a street by a crow; the Community Note clarified that the crow was probably trying to kill it, and the original poster deleted the post.) At times, the disputes can be wildly annoying or pedantic and underscore just how severe a waste of your one life it is to be online at all. I laughed recently at an X post: 'People really log on here to get upset at posts and spend their time writing entire community notes that amount to 'katy perry isn't an astronaut.'' [Read: The perfect pop star for a dumb stunt] The upside, though, is that when anything can be annotated, it feels like less of a big deal or a grand conspiracy when something is. Formal fact-checking programs can feel punitive and draconian, and they give people something to rally against; notes come from peers. This makes receiving one potentially more embarrassing than receiving a traditional fact-check as well; early research has shown that people are likely to delete their misleading posts when they receive Community Notes. The optimistic take on notes-type systems is that they make use of material that already exists and with which everyone is already acquainted. People already correct each other online all the time: On nearly any TikTok in which someone is saying something obviously wrong, the top comment will be from another person pointing this out. It becomes the top comment because other users 'like' it, which bumps it up. I already instinctively look to the comment section whenever I hear something on TikTok and think, That can't be true, right? For better or worse, the idea of letting the crowd decide what needs correcting is a throwback to the era of internet forums, where actually culture got its start. But this era of content moderation will not last forever, just as the previous one didn't. By outright saying that a cultural and political vibe, of sorts, inspired the change, Meta has already suggested as much. We live on the actually internet for now. Whenever the climate shifts—or whenever the heads of the platforms perceive it to shift—we'll find ourselves someplace else. Article originally published at The Atlantic
Atlantic
05-05-2025
- Business
- Atlantic
Fact-Checking Is Out, ‘Community Notes' Are In
One Friday in April, Meta's chief global affairs officer, Joel Kaplan, announced that the process of removing fact-checking from Facebook, Threads, and Instagram was nearly complete. By the following Monday, there would be 'no new fact checks and no fact checkers' working across these platforms, which are used by billions of people globally—no professionals marking disinformation about vaccines or stolen elections. Elon Musk, owner of X—a rival platform with an infamously permissive approach to content moderation— replied to Kaplan, writing, 'Cool.' Meta, then just called Facebook, began its fact-checking program in December 2016, after President Donald Trump was first elected and the social network was criticized for allowing the rampant spread of fake news. The company will still take action against many kinds of problematic content—threats of violence, for example. But it has left the job of patrolling many kinds of misinformation to users themselves. Now, if users are so compelled, they can turn to a Community Notes program, which allows regular people to officially contradict one another's posts with clarifying or corrective supplementary text. A Facebook post stating that the sun has changed color might receive a useful correction, but only if someone decided to write one and submit it for consideration. Almost anyone can sign up for the program (Meta says users must be over 18 and have accounts 'in good standing'), making it, in theory, an egalitarian approach to content moderation. Meta CEO Mark Zuckerberg has called the pivot on misinformation a return to the company's 'roots,' with Facebook and Instagram as sites of 'free expression.' He announced the decision to adopt Community Notes back in January, and explicitly framed the move as a response to the 2024 elections, which he described as a 'cultural tipping point towards once again prioritizing speech.' Less explicitly, Meta's shift to Community Notes is a response to years of being criticized from both sides of the aisle over the company's approach to misinformation. Near the end of his last term, Trump targeted Facebook and other online platforms with an executive order accusing them of 'selective censorship that is harming our national discourse,' and during the Biden administration, Zuckerberg said he was pressured to take down more posts about COVID than he wanted to. Meta's abandonment of traditional fact-checking may be cynical, but misinformation is also an intractable problem. Fact-checking assumes that if you can get a trustworthy source to provide better information, you can save people from believing false claims. But people have different ideas of what makes a trustworthy source, and there are times when people want to believe wrong things. How can you stop them? And, the second question that platforms are now asking themselves: How hard should you try? Community Notes programs—originally invented in 2021 by a team at X, back when it was still called Twitter—are a somewhat perplexing attempt at solving the problem. It seems to rely on a quaint, naive idea of how people behave online: Let's just talk it out! Reasonable debate will prevail! But, to the credit of social-media platforms, the approach is not as starry-eyed as it seems. The chief innovation of Community Notes is that the annotations are generated by consensus among people who might otherwise see things differently. Not every note that is written actually appears under a given post; instead, they are assessed using ' bridging' algorithms, which are meant to 'bridge' divides by accounting for what's called 'diverse positive feedback.' This means that a potential note is valued more highly and is more likely to appear on a post if it is rated 'helpful' by a wide array of people who have demonstrated different biases at other times. The basics of this system have quickly become a new industry standard. Shortly after Meta's announcement about the end of fact-checking, TikTok said that it would be testing its own version of Community Notes, called Footnotes—though unlike Meta and X, TikTok will keep using a formal fact-checking program as well. These tools are 'a good idea and do more good than harm,' Paul Friedl, a researcher at Humboldt University, in Berlin, told me. Friedl co-authored a 2024 paper on decentralized content moderation for Internet Policy Review, which discussed X's Community Notes among other examples, including Reddit's forums and old Usenet messaging threads. A major benefit he and his co-author cited was that these programs may help create a 'culture of responsibility' by encouraging communities 'to reflect, debate, and agree' on the purpose of whatever online space they're using. Platforms certainly have good reasons to embrace the model. The first, according to Friedl, is the cost. Rather than employing fact-checkers around the world, these programs require only a simple algorithm. Users do the work for free. The second is that people like them—they often find the context added to posts by fellow users to be helpful and interesting. The third is politics. For the past decade, platforms—and Meta in particular—have been highly reactive to political events, moving from crisis to crisis and angering critics in the process. When Facebook first started flagging fake news, it was perceived as too little, too late by Democrats and reckless censorship by Republicans. It significantly expanded its fact-checking program in 2020 to deal with rampant misinformation (often spread by Trump) about the coronavirus pandemic and that year's election. From March 1, 2020, to Election Day that year, according to Facebook's self-reporting, the company displayed fact-checking labels on more than 180 million pieces of content. Again, this was perceived as both too much and not enough. With a notes-based system, platforms can sidestep the hassle of public scrutiny over what is or isn't fact-checked and why and cleanly remove themselves from drama. They avoid making contentious decisions, Friedl said, which helps in an effort 'not to lose cultural capital with any user bases.' John Stoll, the recently hired head of news at X, told me something similar about Community Notes. The tool is the 'best solution' to misinformation, he said, because it takes 'a sledgehammer to a black box.' X's program allows users to download all notes and their voting history in enormous spreadsheets. By making moderation visible and collaborative, instead of secretive and unaccountable, he argued, X has discovered how to do things in 'the most equitable, fair, and most pro-free-speech way.' ('Free speech' on X, it should be noted, has also meant platforming white supremacists and other hateful users who were previously banned under Twitter's old rules.) Read: X is a white-supremacist site People across the political spectrum do seem to trust notes more than they do standard misinformation flags. That may be because notes feel more organic and tend to be more detailed. In the 2024 paper, Friedl and his co-author wrote that Community Notes give responsibilities 'to those most intimately aware of the intricacies of specific online communities.' Those people may also be able to work faster than traditional fact-checkers—X claims that notes usually appear in a matter of hours, while a complicated independent fact-check can take days. Yet all of these advantages have their limits. Community Notes is really best suited to nitpicking individual instances of people lying or just being wrong. It cannot counter sophisticated, large-scale disinformation campaigns or penalize repeated bad actors (as the old fact-checking regime did). When Twitter's early version of Community Notes, then called Birdwatch, debuted, the details of the mechanism were made public in a paper that acknowledged another important limitation: The algorithm 'needs some cross-partisan agreement to function,' which may, at times, be impossible to find. If there is no consensus, there are no notes. Musk himself has provided a good case study for this issue. A few Community Notes have vanished from Musk's posts. It's possible that he had them removed—at times, he has seemed to resent the power that X has given its users through the program, suggesting that the system is 'being gamed' and chiding users for citing 'legacy media'—but the disappearances could instead be an algorithmic issue. An influx of either Elon haters or Elon fans could ruin the consensus and the notes' helpfulness ratings, leading them to disappear. (When I asked about this problem, Stoll told me, 'We're, as a company, 100 percent committed to and in love with Community Notes,' but he did not comment on what had happened to the notes removed from Musk's posts.) The early Birdwatch paper also noted that the system might get really, really good at moderating 'trivial topics.' That is the tool's core weakness and its core strength. Notes, because they are written and voted on by people with numerous niche interests and fixations, can appear on anything. While you'll see them on something classically wrong and dangerous, such as conspiracy theories about Barack Obama's birth certificate, you'll also see them on things that are ridiculous and harmless, such as a cute video of a hedgehog. (The caption for a hedgehog video I saw last week suggested that a stumbling hedgehog was being 'helped' across a street by a crow; the Community Note clarified that the crow was probably trying to kill it, and the original poster deleted the post.) At times, the disputes can be wildly annoying or pedantic and underscore just how severe a waste of your one life it is to be online at all. I laughed recently at an X post: 'People really log on here to get upset at posts and spend their time writing entire community notes that amount to 'katy perry isn't an astronaut.'' The upside, though, is that when anything can be annotated, it feels like less of a big deal or a grand conspiracy when something is. Formal fact-checking programs can feel punitive and draconian, and they give people something to rally against; notes come from peers. This makes receiving one potentially more embarrassing than receiving a traditional fact-check as well; early research has shown that people are likely to delete their misleading posts when they receive Community Notes. The optimistic take on notes-type systems is that they make use of material that already exists and with which everyone is already acquainted. People already correct each other online all the time: On nearly any TikTok in which someone is saying something obviously wrong, the top comment will be from another person pointing this out. It becomes the top comment because other users 'like' it, which bumps it up. I already instinctively look to the comment section whenever I hear something on TikTok and think, That can't be true, right? For better or worse, the idea of letting the crowd decide what needs correcting is a throwback to the era of internet forums, where actually culture got its start. But this era of content moderation will not last forever, just as the previous one didn't. By outright saying that a cultural and political vibe, of sorts, inspired the change, Meta has already suggested as much. We live on the actually internet for now. Whenever the climate shifts—or whenever the heads of the platforms perceive it to shift—we'll find ourselves someplace else.
