For X owner Elon Musk, the solution to monitoring misinformation online has been the community, rather than a group of fact checkers. Since buying the social media company formerly known as Twitter in 2022, he’s touted the Community Notes feature as the best way to correct false posts.

That is, until he didn’t like the results.

Musk wrote in a post Thursday that he intends to “fix” Community Notes because it “is increasingly being gamed by governments & legacy media.” He provided no evidence to support his claim.

What apparently set Musk off was information members added in Community Notes correcting posts on X that claimed Volodymyr Zelenskyy, the country’s elected president, had low approval ratings among its citizens.

The Kyiv International Institute of Sociology, or KIIS, published survey results this week, based on February polling, that found that 57% of Ukrainians said they trusted Zelenskyy while 37% said they did not. The polling contradicted President Donald Trump’s claim that Zelenskyy is deeply unpopular in his country.

Tensions between the Trump administration, which includes Musk as a central figure, and the Ukrainian government have escalated over the past week, NBC News reported, before bubbling into public view.

Echoing Kremlin sentiments, Trump has accused Ukraine of starting a war with Russia that actually began when Russian President Vladimir Putin ordered his troops to invade the neighboring country in February 2022.

Senior White House officials met their Russian counterparts in Saudi Arabia on Feb. 18, with the aim of laying the groundwork for peace talks on Ukraine, while excluding Kyiv’s officials and EU representatives from participating in discussions.

Trump has called on Ukraine to hold new elections.

Zelenskyy said he would reject any plan that did not include Ukraine’s involvement. Under its constitution, Ukraine can’t hold elections while it’s at war and under martial law.

In his post Thursday, Musk wrote, “It should be utterly obvious that a Zelensky-controlled poll about his OWN approval is not credible!!” However, there are other available sources.

A consortium that’s been conducting extensive polling in Ukraine since 2014 found “63% of Ukrainians now approve of Zelensky’s performance as president, a notable increase from the previous year,” Joe Stafford, the news and media relations lead at the University of Manchester, wrote in a post Wednesday.

High favorability ratings for Zelenskyy undermine the narrative that Trump and Musk want to tell.

“If Zelensky was actually loved by the people of Ukraine, he would hold an election,” Musk wrote, again without evidence. “He knows he would lose in a landslide, despite having seized control of ALL Ukrainian media, so he canceled the election. In reality, he is despised by the people of Ukraine.”

First introduced by Twitter in 2021 as Birdwatch, and rebranded as Community Notes after Musk’s acquisition the following year, the feature was meant to help the social network combat misinformation and disinformation by enlisting users to flag misleading posts and provide correct information instead.

Community Notes has worked in a manner similar to Wikipedia. Facebook owner Meta, which has been aggressively seeking Trump’s favor in the early days of his second term in the White House, recently announced its own version of Community Notes. Alphabet’s YouTube has been testing a comparable tool since summer 2024.

Neil Johnson, a George Washington University physics professor who studies how misinformation and hate speech spread online, said the Community Notes model is problematic, but not because it can be gamed by large institutions. Rather, crowdsourcing is an inherently “imperfect system” for landing on the truth and is a poor substitute for “formal fact checking,” he said.

“Like any crowd, crowds can be fickle, and crowds can be driven by other interests,” Johnson said. “It’s not a paid person with the job of fact checking.”

Additionally, while Musk has pitched Community Notes as a way to replace fact checkers, a recent study by the Spanish fact-checking nonprofit Maldita showed that many X users still rely on information from professionals. The authors of the study, published earlier in February, looked at more than 1 million notes from Community Notes’ public dataset.

“The evidence from X clearly shows that users rely on the work of fact-checking organizations often” when proposing Community Notes, they wrote.

Neither Musk nor a representative from X responded to CNBC’s requests for comment.

Musk’s latest comments on Community Notes mark a sharp contrast to how he’s discussed the service in the past, and underscore the lengths to which he’s willing to go in pursuit of Trump’s agenda.

When talking about Community Notes in earlier posts, Musk has said that it can’t be manipulated by him or anyone else.

“The system is completely decentralized and open source, both code and data. Any manipulation would show up like a neon sore thumb!” Musk wrote in a post Dec. 30. “No one at X, including me, has any editorial control.”

Musk has also acknowledged in past posts that he is not immune from being corrected in Community Notes.

And Community Notes isn’t the only technology in Musk’s portfolio that could present problems by responding in ways he may not like. There’s also Grok, the artificial intelligence chatbot that is owned by Musk’s startup xAI and is used on X.

Fortune published a story in January about the many negative responses Grok provides when users ask if Musk is a good person. Grok’s reasons for saying he isn’t a good person include environmental hazards from SpaceX, Musk’s “erratic” management style and his political views, Fortune reported, citing Grok responses. Futurism published a similarly themed piece in December, with the headline “Elon Musk’s Grok AI blasts Elon Musk as huge spreader of misinformation.”

Musk calls Grok a “maximally truth-seeking” AI that is also “anti-woke.” Earlier this week, xAI introduced its latest AI model, Grok 3, claiming it can outperform offerings from OpenAI and China’s DeepSeek based on early testing, which included standardized tests on math, science and coding. 

Musk did admit during the demo that the model isn’t perfect.

“We should emphasize that this is kind of a beta, meaning that you should expect some imperfections,” Musk said. “But we will improve it rapidly, almost every day.”

WATCH: Elon Musk’s xAI unveils new Grok model

When Apple revealed the Vision Pro in 2023, it called the $3,500 headset its next “major platform.” Two years later, and a year after going on sale, the device is thin on apps. 

Apple doesn’t regularly release stats on the number of Vision Pro apps that are available, and it’s hard to tell how many new apps come out in any given month. According to consultancy AppFigures, which tracks Apple’s platforms, the number of new Vision Pro apps has declined every month since the device hit the market in February 2024.

When Apple unveiled the Vision Pro, executives said that developers would be able to create new experiences that weren’t possible with traditional computers. But so far, top developers remain mostly focused elsewhere, and major tech companies like Google, Meta and Netflix have yet to release their most important apps for the headset.

Many of the new apps and ideas for the Vision Pro are coming from independent developers, hacking on the weekends while holding down day jobs. 

One person in the indie camp is Adam Roszyk, a programmer in Poland who has created 17 Vision Pro apps since the headset was first released.

For $4, Roszyk’s Night Vision app lets a Vision Pro user tap the depth-sensing cameras of the device to see objects in the dark. If you spend $5, you can perform a chore in a Luigi’s Mansion-like video game using the app Vacuume, which overlays virtual coins on your floor that you can vacuum up, along with any real dirt or dust. And for $6, Roszyk’s app Scan Export lets users create a 3D digital scan of an entire building just by walking around, a useful tool for those in construction or real estate. 

“We are still early, and we don’t really know how it can be really useful in your life,” Roszyk said. “There’s so many different ideas that just come to your mind.”

Roszyk continues to work on Vision Pro apps because he said he believes “spatial computing” — Apple’s preferred terminology for headset and glasses technology that can integrate 3D objects with the world around them — will be the next big platform. Roszyk is betting that developing apps now can put him in prime position when more people are walking around with a Vision Pro or, perhaps some day, lightweight glasses. 

“This type of computing is the future,” Roszyk said. “I would definitely compare it to the first iPhones.”

Roszyk’s efforts have made him money, but not enough for Vision Pro development to become his full-time job. His 17 apps have cleared about $4,000 on the App Store in the last three months. That number is growing as he releases more apps and more people find out about them, Roszyk said. 

Apple updated its most recent Vision Pro app count in August, with CEO Tim Cook telling investors on an earnings call that the platform had 2,500 apps. That number covers fully immersive apps that overlay virtual objects over the real world as well as 2D apps with some spatial components.

By AppFigures’ count, less than 1,900 of these apps remained active at the end of January.

Apple declined to comment.

Rival Meta in 2023 said that it had 500 apps in its Quest store, and the company last year said that number had multiplied by 10.

The Quest 3S, which has many of the same features as the Vision Pro, starts at $300. Meta also sold millions of its predecessors in recent years. While Meta hasn’t revealed how many users it has, its Meta Quest app was downloaded about 6 million times in 2024, according to AppFigures data, a useful proxy because users need to download the app in order to set up the headset.

There are also about 1.5 million Vision Pro apps that are ported versions of iPhone and iPad apps. Apple automatically ports iPhone and iPad apps to the Vision Pro when they’re uploaded, but companies can decline. Those apps can be used inside the headset but appear as 2D flat screens. Meta started to emulate that strategy last year with 2D Android apps for Quest, but the company doesn’t have the same library of millions of existing mobile apps.

Apple doesn’t publish Vision Pro sales, but one estimate from IDC suggests fewer than 1 million devices have been sold.

Some services like Netflix and YouTube, and game streaming services like Nvidia GeForce Now can be accessed through the Apple Vision Pro’s browser. And existing apps often receive updates that introduce a spatial mode, such as the NBA scores app, which recently got an experimental feature that allows users to watch a live basketball game as if the players were miniature figurines on a table.

Apple Arcade, a monthly game subscription from Apple, does require that its titles support the Vision Pro in addition to iPhones and iPads. Apple Arcade developers are paid by Apple and their apps are free to subscribers.

Although many of those games are 2D, some are exclusive to the Vision Pro. In January, Apple released Gears & Goo, a Vision Pro app that enables the player to control an army of goofy frog-like characters on a table in the real world.

Meanwhile, Meta is actively courting VR developers with a promise that they can make money. Meta in January said that its payment volume for Quest headsets rose by 12% last year, although it didn’t cite a total number. Meta has also said it has 200 apps that have made more than $1 million through software sales.

A year after the iPhone App Store was launched in 2008, Apple was crowing about the platform having 50 million customers, 2 billion downloads and 85,000 apps. Apple regularly told investors and developers how much money it had paid from App Store sales — it hasn’t released any similar stat for the Vision Pro.

Many in the VR industry hoped Apple’s entry would kick off a boom like the iPhone did for mobile apps, creating fortunes as millions of users sought to fill their new devices with fresh software.

“My assumption back then was whatever Apple releases might be in that final form, so it’s a good idea to be ready as early as possible,” said Nikhil Jacob, who runs Vision Uni, which publishes content about developing apps for the Vision Pro. “But my assumption there ended up being wrong.”

Jacob said he believes that an app developer ecosystem for the Vision Pro will take a lot longer to build out than it did for the iPhone because key pieces are missing. Jacob hopes Apple improves the Vision Pro app store to help users find new apps.

The slow uptake, due largely to the high price tag, has led some to worry that VR and its related technologies are once again entering a lull.

“Winter has come,” said Jarrett Webb, who develops headset apps for Argodesign, a software consultancy. “Even Apple couldn’t produce a winner.”

Still, some optimism remains among Vision Pro developers. 

They say that Apple’s hardware is solid, the company’s developer tools are improving, and that the Vision Pro lays the groundwork for future software and hardware updates. It also helps that Vision Pro owners still seem to be excited to try out new apps.

Apple’s entry into the headset market, combined with Google’s recent announcement of its own Android XR platform, as well as Meta’s billions of dollars of investment signals that there will be a market for VR content, said John Gearty, who worked on the Vision Pro at Apple and is the founder of PulseJet Studios, a VR production house focusing on music. Gearty is hoping for steady growth from the market, but he has tempered his expectations. 

“I don’t think it’s ever going to be hockey stick growth,” he said.

Apple has not said if it will update the Vision Pro. According to analysts, the company is working on a successor. Developers want it to be lighter and less expensive. They welcome any improvements that would get it on more faces.

“Over time, everything gets better, and it too will have its course of getting better and better,” Cook told The Wall Street Journal in October. “I think it’s just arguably a success today from an ecosystem-being-built-out point of view.”

— CNBC’s Jonathan Vanian contributed to this report.

Chinese artificial intelligence lab DeepSeek roiled markets in January, setting off a massive tech and semiconductor selloff after unveiling AI models that it said were cheaper and more efficient than American ones. 

But the underlying fears and breakthroughs that sparked the selling go much deeper than one AI startup. Silicon Valley is now reckoning with a technique in AI development called distillation, one that could upend the AI leaderboard. 

Distillation is a process of extracting knowledge from a larger AI model to create a smaller one. It can allow a small team with virtually no resources to make an advanced model.

A leading tech company invests years and millions of dollars developing a top-tier model from scratch. Then a smaller team such as DeepSeek swoops in and trains its own, more specialized model by asking the larger “teacher” model questions. The process creates a new model that’s nearly as capable as the big company’s model but trains more quickly and efficiently. 

“This distillation technique is just so extremely powerful and so extremely cheap, and it’s just available to anyone,” said Databricks CEO Ali Ghodsi, adding that he expects to see innovation when it comes to how large language models, or LLMs, are built. “We’re going to see so much competition for LLMs. That’s what’s going to happen in this new era we’re entering.” 

Distillation is now enabling less-capitalized startups and research labs to compete at the cutting edge faster than ever before.

Using this technique, researchers at Berkeley said, they recreated OpenAI’s reasoning model for $450 in 19 hours last month. Soon after, researchers at Stanford and the University of Washington created their own reasoning model in just 26 minutes, using less than $50 in compute credits, they said. The startup Hugging Face recreated OpenAI’s newest and flashiest feature, Deep Research, as a 24-hour coding challenge. 

DeepSeek didn’t invent distillation, but it woke up the AI world to its disruptive potential. It also ushered in the rise of a new open-source order — a belief that transparency and accessibility drive innovation faster than closed-door research.

“Open source always wins in the tech industry,” said Arvind Jain, CEO of Glean, which makes an AI-powered search engine for enterprises. “You cannot beat the momentum that a successful open-source project is able to actually generate.” 

OpenAI itself has walked back its closed-source strategy in the wake of DeepSeek’s accomplishment.

“Personally I think we have been on the wrong side of history here and need to figure out a different open-source strategy,” OpenAI CEO Sam Altman wrote in a post on Reddit on Jan. 31. 

The combination of distillation’s newfound traction and open source’s rise in popularity is completely altering the competitive dynamics in AI. 

Watch the video to learn more.