Meta sacrifices a heap of money at the altar of AI

6 hours ago 3

Mark Zuckerberg announced in April that the company would make huge capital expenditures in the coming year to keep up in the race to develop cutting-edge artificial intelligence. He made good on that promise last week with a $15bn “AI superintelligence” team that would feature reported nine-figure salaries and a 49% investment in Scale AI. Meta also hired Scale’s 28-year-old founder, Alexandr Wang, a former roommate of OpenAI’s Sam Altman.

Before Meta’s investment, Scale counted most of the major players in AI among its clients, and some of them were less than thrilled with the development. Bloomberg puts it succinctly: Scale AI’s Wang Brings to Meta Knowledge of What Everyone Else is Doing. Google, Scale’s largest customer, got scared. The tech giant told the startup that their working relationship would end in response to the deal, Reuters reported on Friday.

My colleague Robert Booth has more:

One Silicon Valley analyst described the announcement as the action of “a wartime CEO”.

Superintelligence is described as a type of AI that can perform better than humans at all tasks. Currently AI cannot reach the same level as humans in all tasks, a state known as artificial general intelligence (AGI).

Meta’s attempt to leapfrog the current state of progress and target superintelligence is seen by observers as an attempt by the company to regain the initiative over AI after significant advances by competitors including Sam Altman’s OpenAI and Google and after Meta’s huge investment in the concept of the Metaverse flopped.

The financial magnitude of the move may seem like taking command of the AI race, but Meta is playing catchup. Its most recent AI models are less capable than those of rivals. In April, the company published a model designed with features tailored to gaming a popular benchmark, a desperate move. With its mammoth investment and the formation of a new team, particularly one helmed by a key industry player, the tech giant is buying its way in.

Read more about Meta’s superintelligence team here.

The week in AI news

Wikipedia, vital to AI, takes pains to highlight human contributions

Last week, Wikipedia halted a test of a new AI-powered article summarizer in response to a backlash from its editors, the volunteers who contribute new information to the online encyclopedia’s articles and evaluate the quality of others’ additions. The site had introduced summaries of articles, generated by artificial intelligence, that would appear at the top of those pages above the introduction to the subject, written by humans. The test ran for two weeks for about 10% users of the mobile version of Wikipedia who opted into it.

The feedback was, in the words of Wikipedia’s nonprofit parent, “strongly negative”. One editor wrote in the public forum: “A truly ghastly idea.” Another: “Keep AI out of Wikipedia.” And: “I am going to join the chorus of editors saying that this is a uniquely bad idea.” Some argued in favor of the introduction of summaries by AI, making the point that technical articles on Wikipedia prioritize depth and complexity over accessibility. Most, though, argued that any hallucinations and mistakes would degrade Wikipedia’s reputation of reliability.

The general sentiment among editors read thusly: “Yuck”.

“We’re trying to figure out the balance between AI and Wikipedia. There are AI integrations that do happen on Wikipedia but are very minimal. Little things that make it easier for editors to do their work or readers to get the content they need,” said Pacita Rudder, executive director of Wikimedia New York City, a local chapter affiliated with the foundation. “‘Where’s the line between human output and robotic output?’ is a debate that’s constantly happening within the Wikipedia community. There are some people who are for it, some against it, but as with anything with Wikipedia, we have to come to a consensus.”

Contrast Wikipedia’s deferential response with Reddit’s. The social network began charging for access to its application programming interface in 2023. Reddit’s library of posts is extremely valuable to AI companies, which need enormous collections of text written by contemporary human beings to train large language models like ChatGPT. Reddit execs intended to extract money from AI companies using the company’s library. They succeeded. They also rendered many users’ experiences worse by crippling popular extensions. In response, users, particularly forums’ volunteer moderators, revolted en masse, blacking out forums with tens of millions of members and demanding reversion. Executives refused. They chose dollars from AI companies over their users’ concerns, the opposite of Wikipedia. A coda to the story, though: two years later, Reddit largely functions as before.

Several days after Wikipedia rolled back its AI test, the Wikimedia Foundation hosted an in-person “edit-a-thon” at the United Nations, an event dedicated to creating and updating pages about the history of the international body, complete with a work list. The conclave of information custodians – a 50/50 balance of new and experienced editors, according to Rudder – was part of the UN’s Open Source Week and occurred alongside an open source hackathon. It was in-person only.

The edit-a-thon stands in contrast to the stereotypical tableau of curating Wikipedia articles: a single person online, communicating sterilely with others via a glaring white screen. At the event, though, editors convened in groups of five around a single laptop to converse excitedly and collaborate in as many languages throughout a large, buzzing auditorium. Italian and Sri Lankan envoys lionized the work of the editors in speeches, as well as the open source community at large. An executive from the Wikimedia Foundation thanked them for their work, as did the UN under-secretary-general for digital and emerging technologies. It was an energetic scene.

Asked why she felt it was important to host an in-person-only event, Rudder said: “It’s community, right? There’s this misconception that a lot of Wikipedia editors are alone in their basements, typing away, but that’s not the case,” said Rudder. Wikimedia NYC hosted the event at the UN. Rudder said the chapter’s role is to bring editors and members of the online community together. She said she felt most energized by meeting new editors.

As with Reddit, Wikipedia’s huge text corpus is part of the foundational training data for most, if not all, large language models. And like the white alien company, Wikipedia charges for smooth access to it, though less than Reddit, Rudder said. The organization launched Wikimedia Enterprise in 2021 with an eye towards corporate customers. Reddit is a for-profit business, whereas the Wikimedia Foundation is a non-profit. Wikipedia’s data is free, but using it easily may cost you.

Unlike Reddit, Wikipedia is emphasizing the human labor necessary to maintain its site and the content there by rolling back features in response to community outcry and convening editors in person.

skip past newsletter promotion

The tech angle on the Los Angeles anti-deportation protests

Apple’s walled garden turns from Eden to a pen

A week ago, Apple hosted its annual developers’ conference, WWDC. The biggest announcement was a redesign of the iPhone Operating System (iOS) nicknamed “liquid glass”. Virtual buttons on iPhone screens will appear transparent, with the images behind them passing through with a slight warp like a magnifying glass. The aesthetic recalls the colorful Macs of old, housed in translucent, whimsical plastic.

The refreshed design is pretty! It is also boring. Machines are fabricating entire movies and predicting hurricanes. Who cares what color my phone’s controls are?

There were other updates, of course. Apple is adding live translation to FaceTime calls, for instance. I can imagine many calls with distant family and friends that will run smoother with that feature. How nice.

On the whole, though, the updates failed to inspire. For many years, updates to iPhones have seemed far more incremental than, for example, the addition of speedy 3G connectivity in 2008 with the second generation. Apple’s updates have failed to keep pace with the freewheeling Android ecosystem. I met a venture capitalist in San Francisco in 2019 who did not know what version of the iPhone he was holding in his hand as we spoke.

Apple showcased little in the way of new AI features, unlike its peers. Google and Samsung are integrating AI into their mobile experiences at a breakneck pace. Case in point: live translation is a feature that has been part of the Android operating system for several years. Google’s Pixel Bud headphones can translate the speech of a person talking to you, which Airpods cannot do. Google seems to release a fascinating, if sometimes bizarre, AI feature every week. As recently as Friday, the Verge reported: Google says it is testing Audio Overviews for certain Search queries, available for Labs users in English in the US.

After the demonstration, Gizmodo argued: Apple Knows AI Isn’t What People Really Want, but It Can’t Say That. Even if that is true, though – and ChatGPT’s 500m monthly users might disagree – the announcements showed little in the way of flair. If Apple wants users to stay within the iPhone’s ecosystem, it needs to offer them something equally appealing and enthralling or even more so.

Google pushed a new, updated version of the operating system and last week. Samsung teased new features of the next version of its folding phone, per Android Central. Did you hear about them? It is axiomatic in tech journalism that stories about Apple and iPhones will always garner more clicks than stories about Android phones, though a far higher percentage of the world’s phones use Android. I have many guesses as to why that is, but it is a topic for another day. I only mention it because many fewer people spent the week discussing the Android updates as did Apple’s Liquid Glass interface. Apple’s change amounts to only a cosmetic upgrade – and it’s debatable whether it will be better at all – but it inspired many more jokes.

Samsung’s seventh version of its folding phone, while novel, begs the question: as I complain about new iPhones’ lack of thrills, what do I want my phone to do? I do not need a folding phone. I had one, a Motorola Razr, and it did facilitate vital communication; however, that did not arise from the form factor.

Google’s new AI features pose the same question. There may not be consensus on what the majority of users want from an AI-enhanced phone, but if you are not trying something new with technology, you are in decline. It doesn’t feel like Apple is offering surprising features that I would not have thought of. I may not need an audio summary feature for my Google search results, but I am intrigued by it. Also, what a gift to the visually impaired! Siri, by contrast, still struggles to control Spotify and botches notification summaries.

I have used an iPhone as my primary device for about 15 years now. This inflection point would be the moment to switch to Android to try out its shiny toybox of new features, but I am enclosed in Apple’s ecosystem of hardware and apps, a walled garden that is starting to feel less like a bountiful Eden and more like a pen.