Welcome to AI Decoded, Fast Company’s weekly newsletter that breaks down the most important news in the world of AI. You can sign up to receive this newsletter every week here.
A cooler-headed look at Apple’s troubled AI story
Apple said at its WWDC developer conference Monday that the much-hyped Apple Intelligence features it announced at last year’s event are still not ready to ship—and likely won’t be until 2026. Now, the company is living with a media narrative that it’s failed, fallen behind, stuck in neutral, or in retreat on AI.
To be clear, Apple’s various operating systems already include AI-powered features, from photo cleanup to call screening. And the company did deliver some of the Apple Intelligence tools it promised last year. But those were first-generation large language model functions like text message summarization and writing assistance—unambitious by generative AI standards and not especially useful.
What got everyone excited at WWDC 2024 were the advanced, personalized AI features that allow Siri to leverage a user’s personal data and carry out actions on their behalf. Some examples:
- A user could say, “Hey Siri, when does Mom’s flight land?” and Siri—having access to the user’s email, contacts, calendar, and files—would know the answer.
- Siri would also be able to “see” what’s on the user’s screen, so it could, for example, read an address from a text message and add it to a contact card.
- A user could tell Siri to take action within an app, such as: “Siri, take the red-eye out of this photo,” or “Send the email I drafted to Judy and Carmen.”
Cool stuff, useful, and things Apple is uniquely positioned to offer. So what happened?
Apple software chief Craig Federighi and head of marketing Greg “Joz” Joswiak offered a fuller explanation during a round of media interviews Tuesday. According to them, Apple had a working “version one” of a new Siri architecture before announcing the personal AI features at WWDC 2024. (In other words: No, the demos weren’t vaporware.) The Siri team believed they could continue polishing the product until it performed reliably enough for a late 2024 release.
But Federighi explained in an ">interview with TechRadar that the system continued to generate too many unreliable results in internal testing. When a year-end release became unrealistic, the team aimed for spring 2025. But by then, the system was still inconsistent. That’s when the Siri team concluded it would need to create a “version two” of the new Siri architecture—one that “extends across the entire Siri experience.” Until that reworked Siri is performing reliably in-house, Federighi said, the company won’t speculate on a public release date.
Federighi also argued that Apple is trying to do something with personalized AI that nobody else has really done. There’s some truth to that—especially if we limit the field to mobile-based AI assistants, and when we’re talking about doing it at Apple’s scale. Apple may look like an AI laggard now, but it would look far worse if it shipped something that didn’t work or wasn’t useful.
Still, it’s not accurate to say Apple is the only one in the game. In May, Google announced a suite of personalized AI features that will begin rolling out this summer. Its Gemini assistant can use personal data in email to offer summaries, draft messages, provide reminders, and deliver context-aware insights. Gemini Live can ">analyze objects in the camera view and on the screen. And the “agent mode” in the Gemini app, built on Project Mariner, can perform multi-step tasks for users both online and within apps.
Apple’s big mistake, of course, was announcing personal AI features in 2024 that it couldn’t deliver in 2024—or even 2025. That misstep might not have happened if this were just another set of traditional software updates. But we’re not in that era anymore. Apple is now operating in the realm of probabilistic computing—the domain of AI models. The usual rules for predicting software maturity and release timelines may no longer apply.
Meta to rebrand itself “MetaGPT” (just kidding)
Okay, forgive the Onion-esque headline, but it holds a seed of truth. Once again, Mark Zuckerberg is spending billions to push his company to the front of the line in the next big thing in consumer tech—this time, it’s generative AI. (A few years ago, Facebook renamed itself “Meta” when it bet that the metaverse was the next big thing.)
Now, The New York Times and Bloomberg report that the social media giant—after seeing its Llama models lag behind those from OpenAI, Google, and Anthropic—is launching a new generative AI lab with ambitions of achieving “superintelligence.” (OpenAI, Anthropic, and others are pursuing “artificial general intelligence,” where AI matches or exceeds human abilities across a wide range of tasks. Superintelligence refers to AI that vastly outperforms humans at many tasks.)
Zuckerberg has reportedly agreed to buy a 49% stake in Alexander Wang’s Scale AI—a company that specializes in human-labeled and synthetic training data for AI models—for nearly $15 billion. This appears to be another “acquihire,” where Meta’s primary interest is Wang himself. The 28-year-old will lead the new superintelligence group at Meta. (Microsoft made a similar move when it acquired Inflection AI and brought on its founder and CEO, Mustafa Suleyman.)
Meta is also offering “seven to nine figure salaries” to attract top researchers to the new lab.
The company already has a major AI figure in Yann LeCun, who leads its core AI research group. But The Times reports that Meta has experienced significant internal friction over how to approach the development and deployment of AI models. The company’s focused AI efforts began in 2013 after it failed in a bid to acquire DeepMind, which ultimately went to Google.
Unlike OpenAI and Anthropic, Meta doesn’t rely on selling AI models or apps. Its primary business is advertising. Thanks to its massive ad revenue, Meta can afford to open-source its AI models, effectively giving developers free access in hopes of flooding the ecosystem and becoming dominant through ubiquity. But if Zuckerberg succeeds in getting Meta to the front of the superintelligence race, that strategy could shift. The company might begin locking down its models and charging for access—just like Google, OpenAI, and Anthropic.
Hollywood is going after Midjourney, and it could be just the start
Over the past couple of years, a range of content owners—news organizations, authors, artists, comics, and record labels—have filed lawsuits against generative AI companies. Their claim: these companies used copyrighted content, usually scraped from the internet, to train AI models without permission or compensation. In response, AI firms often argue that their use of such material falls under the “fair use” provision of the Copyright Act. OpenAI and Microsoft, for example, have made that argument in an ongoing copyright case brought by The New York Times.
Until now, the big Hollywood studios had mostly stayed on the sidelines. That changed today when the two largest, Universal and Disney, filed a joint copyright infringement lawsuit against Midjourney, one of the original AI image generators. (Midjourney is expected to roll out a new image-to-video feature this month, which may have prompted the legal action.) The studios have been under growing pressure from artists and writers to challenge AI labs over how training data is sourced.
The Disney and Universal suit may be the opening salvo in a much broader conflict between Hollywood and the AI industry. Why start with Midjourney, a relatively small player, when far larger and wealthier AI companies have used the same kinds of scraped, copyrighted data to train their models?
As The New York Times notes, the lawsuit appears to set the stage for something bigger. The language goes beyond a simple dispute between two companies. The plaintiffs argue that the use of copyrighted training data “threatens to upend the bedrock incentives of U.S. copyright law that drive American leadership in movies, television and other creative arts.”
Stay tuned.
More AI coverage from Fast Company:
- Databricks’ new ‘One’ dashboard brings AI to the business class
- This corny ‘conservative credit card’ ad signals a very scary future for AI
- OpenAI and Anthropic are getting cozier with government
- Teaching AI isn’t enough—we need to teach wisdom, too
Want exclusive reporting and trend analysis on technology, business innovation, future of work, and design? Sign up for Fast Company Premium.
Accedi per aggiungere un commento
Altri post in questo gruppo
As AI car crashes go, the recent publishing of a
The Bible is now on TikTok, vlog-style.
Picture David—yes, that David, of Goliath fame—with an iPhone and influencer energy. “Asked the guy to film it, but guess what? The camera froze,”
“In three years,” a fellow tech executive recently told me with serene confidence, “Everyone will be able to make a full-length movie in AI, totally personalized for them, by just typing up a few
In 1999, I got to work on a literally once-in-a-lifetime project. As the 20th century was wrapping up, the magazine where I worked declared the personal computer the most important invention of th
Right-wing conspiracy theorists once believed the government was using drones to surveil its own citizens. President Donald Trump actually did.
During ongoing demonstrations in Los Angel
In summer 2020, former Evernote CEO Phil Libin launched Mmhmm, a soft
From family photos in the cloud to email archives and social media accounts, the digital lives of Americans are extensive and growing.
According to recent studies by the password managem
