After months of obsessing by tech media, Apple’s generative AI features are (almost) here. In a prerecorded keynote at its Worldwide Developers’ Conference (WWDC) on Monday, company executives ran through a series of features like proofreading and changing the tone of text, recording and transcribing phone calls, a more capable Siri voice assistant, and an emoji generator which all fell under the label of “Apple Intelligence” — its branding term for generative AI. But none of them were anything to get too excited about.

The biggest takeaway from Apple’s keynote is that the air continues to flow out of the generative AI bubble. Last month, Google and OpenAI had their own lackluster showcases to demonstrate a series of evolutionary and even mundane features that were hard for anyone not steeped in AI boosterism to get excited about. One of the standout subpar moments of the Google event was when the company showed off how its Gemini chatbot could find a user’s license plate number in the Photos app as if it was a huge deal. Apple virtually recreated it, but with a driver’s license number instead.

AI hype is over. AI exhaustion is setting in.

Google and OpenAI’s latest showcases suggest the AI bubble’s days are numbered

DisconnectParis Marx

Unlike what OpenAI and Google were doing last year, Apple clearly isn’t trying to make people think its generative AI features are going to change the world. If we weren’t in this moment of AI hype driven by investor exuberance rather than tangible technological progress, these features would just be like any other additions to Apple’s operating systems. But they have to be singled out and given some extra fanfare because that’s usually what helps boost a company’s share price these days — or at least protects it from sliding.

After all the build up, Apple’s stock dropped in after market trading but jumped on Tuesday. The company clearly has a strategy to use its Apple Intelligence features to try to address some of the deeper problems with its business. It just remains to see if it will work, and if some of those generative AI features it’s banking on to deliver a boost might come back to bite it.

Apple’s ambitions still face challenges

Last month, Apple was in the crosshairs of artists when it released an advertisement for the new iPad Pro that showed a whole range of creative tools being destroyed to give way to the thin black slab of glass and metal. The company meant to illustrate how much creative work it believed could be done on the device, but after over a year of concern about how tech companies are deploying generative AI to take work away from artists, it sparked an uproar. As Apple rolls out its text and image generation features, the company risks more creative ire.

The image generator built into Apple Intelligence contains the strange and glossy aesthetic we’ve come to expect from AI-generated visuals. Not to mention that the use cases the feature is supposed to make possible — like being able to send personalized generated images in text messages — seem wholly unnecessary. But the worst part is that Apple isn’t being open about where it’s getting the data that powers its models.

In a news release, the company said it used licensed data and “publicly available data collected by our web-crawler,” with the option for publishers to opt out of having their content gathered as training data. It also assured users that their personal data and interactions are not being used. But the company didn’t say whether copyrighted works were absorbed into its training data without creators’ permission, suggesting it likely followed similar methods as other tech companies. Just days ago, Chamber of Progress launched a campaign to defend the legality of tech companies using copyrighted works in AI training data. Apple is a member of that lobby group.

On top of the potential anger from publishers and artists, Apple could face a public backlash once people actually get their hands on its generative AI features. After Google’s I/O event in May, it was pilloried for the terrible AI Overviews it enabled on Search that gave out an endless stream of incorrect and even dangerous answers. We’ve yet to see whether Siri will have its own controversial recommendations to make or if Apple’s built-in image generator will start spitting out offensive AI slop like so many others. That will be tested in the coming months.

Privacy is also still an open question, but the one it may be least likely to face criticism on. Apple has long presented itself as being uniquely committed to protecting user privacy, even as some groups have called those claims into question for not being as solid as the company suggests. The company’s AI features have previously done a lot of their processing directly on the device, but with generative AI it will be sending some tasks to servers it says will be highly secure, so much that it’s calling them “Private Cloud Compute.”

After looking over Apple’s breakdown of the system, cryptographer and Johns Hopkins University professor Matthew Green wrote on Twitter that “if you gave an excellent team a huge pile of money and told them to build the best ‘private’ cloud in the world, it would probably look like this.” Even with that praise, he warned there will still be a lot of vulnerabilities that could be difficult for researchers for detect. Apple also still hasn’t said whether users will be able to fully opt out of the feature if they choose to. But those protections only apply to Apple Intelligence. If a user wants to use the ChatGPT integration, almost all of that goes out the window and users must agree to share data with OpenAI on far less secure terms.

AI to drive iPhone sales

One thing you can say for Apple’s generative AI plans is that it does seem to have better curated the use cases it’s targeting instead of trying to throw everything at the wall, like some of its competitors. That doesn’t eliminate the broader problems that critics have been identifying with generative AI for more than a year, but the framing of the features does signal how Apple is hoping to use them.

The company isn’t positioning Apple Intelligence as a new subscription that users can sign up for to boost its digital services revenue. Instead, it’s just another set of features made available to people who buy its hardware — like its suite of office software. That not only further dispels the notion that generative AI is some groundbreaking development, but shows Apple is positioning the features to help it drive hardware sales — and specifically a new cycle of iPhone upgrades.

The company has been struggling with plateauing iPhone unit sales for years. It increased sales revenue by hiking prices through the introduction of its iPhone X and later iPhone Pro models, but even that stopped working last year when it had four consecutive quarters of declining revenue. Recently, iPhone sales in China saw a recovery, but that was driven by deep discounts.

Apple Intelligence will be limited to iPads and Macs with M1 chips, meaning certain models released in the last few years, but that’s even more constrained for iPhones. The only existing models that will be able to run the generative AI features when they’re released are the premium 15 Pro and Pro Max models that came out last fall, meaning the WWDC keynote was essentially a message to consumers that they’d better get ready to upgrade if they want the latest and greatest AI tools on their phone.

Roundup: Apple wants to hike iPhone prices again

Read to the end for an important statement about X… or Twitter?

DisconnectParis Marx

Apple Intelligence will be the central part of a big push to drive an iPhone upgrade supercycle in the fall to deliver good financial results for the fiscal year. Rumors suggest Apple is even planning to launch a new model alongside the iPhone 16 series that’s tentatively named the iPhone Slim and will be even more expensive than the Pro Max.

Apple Intelligence is not the company’s “next big thing.” After the Vision Pro bombed and the car project was canceled, generative AI may help drive some additional sales in its existing hardware lineup, but Apple still seems as paralyzed by its success as it’s been for some time. It’s lucky chatbots and image generators aren’t the pivotal moment Sam Altman wanted us to believe, or the company might have really been in trouble.

UPDATE (June 11, 2023): Clarified Apple’s share price movements after markets opened.

Can you remember an Apple product unveiling that went as bad as the Vision Pro? The company known for the iPhone and the Mac had a big task: to convince people to stick a computer on their faces for hours on end, experience the world through cameras and screens, isolate themselves from anyone around them, and pay $3,500 for the privilege. The awkward photos of Tim Cook with the thing on his face in Vanity Fair certainly didn’t help convince people it was a future to embrace.

In the lead up to June 2023 unveiling, the New York Times reported on the internal dissent over the company’s decision to release the product and even Bloomberg’s Apple hype man Mark Gurman had to admit that Cook had mismanaged the project. There were plenty of reasons to believe it was going to be a dud, but there were still a sizeable number of people not willing to write off Apple’s next big thing. After all, this was Apple; one of the most valuable companies in the world that shaped personal computing with the Mac then led the mobile revolution with the iPhone. How could it get a major product release that it had poured $10 billion into so wrong? Its executives must’ve seen something us lowly consumers were missing.

In the days after the Vision Pro’s release in February, there was an initial wave of hype as the early adopters got their hands on it. Videos proliferated of people using it in cars, on streets, and in planes — mainly for views, but it still got the discourse flowing. Even some critics came up with bad justifications to get their hands on one. But that didn’t change the fact it was a bad product that imagined an anti-social future that other companies have tried and failed to foist on the public many times before. And in an all too rare moment, many tech reviewers agreed.

Apple’s Vision Pro headset deserves to be ridiculed

Tech companies want us isolated and constantly staring at screens because it drives profit

DisconnectParis Marx

While complimenting the theoretical possibilities the product offered, it was impossible to ignore the many drawbacks on everything from weight to eye strain and headaches to the fact the product just didn’t offer many compelling use cases. Joanna Stern at the Wall Street Journal opined that the Vision Pro was “the best mixed-reality headset I’ve ever tried,” but that companies like Apple know “these aren’t really the devices we want.” At The Verge, Nilay Patel was even more damning, writing that the headset couldn’t “overcome the inherent nature of cameras and display” and “may have inadvertently revealed that some of these core ideas are actually dead ends.” In other words, the Vision Pro wasn’t a first step toward an augmented reality future but a moment to accept that vision would never be properly realized.

The concept of the Vision Pro might have excited some developers in an Apple lab in Cupertino, but most people are never going to use computers in the way those engineers and their managers imagined. Before the 14-day return window closed, there were plenty of reports of people heading back to the Apple store or popping their headsets back in the post to get their money back. They bought into the hype, they tried the new thing, and they ultimately realized it wasn’t worth it. That’s now being reflected in the sales numbers.

Apple initially had a sales target of 3 million Vision Pro units in its first year, but slowly revised that number down to 900,000. When the product was released, estimates pegged initial sales at around 200,000 units, though it’s not clear how many of those were returned. Even people who kept their devices have recently been sharing on social media that they rarely use them anymore. It was no surprise on Tuesday when Apple analyst Ming-Chi Kuo reported Apple had slashed Vision Pro production even before its international launch, expecting to sell as little as 400,000 units this year. Kuo also suggested a cheaper version had been pushed beyond 2025, if the company makes one at all.

Apple’s Vision Pro lacks any real vision

The company’s headset exists to placate investors, not serve users’ needs

DisconnectParis Marx

For some companies, selling 400,000 units would be a major achievement. But for a company like Apple that sells well over 200 million iPhones every year, along with tens of millions of Macs and iPads, it’s nowhere near the success they need it to be to justify the resources that went into it. And it’s looking ever more likely it will never get there.

The power of tech criticism

The Vision Pro is just one in a long line of tech failures that includes everything from Juicero to the metaverse to the crypto financial system to the self-driving cars that were supposed to be ubiquitous by now. There’s no doubt the tech industry has upended many parts of our lives by getting us to adopt their digital technologies en masse — sometimes by choice; sometimes by force. But the times the industry was wrong are often overlooked in the narrative of tech’s supposed inevitability.

When a major tech company throws a new product into the world, it’s all too common that people immediately assume it’s something we must accept or even try to justify why it should exist (and succeed) regardless of how good it is. We’ve seen a lot of that over the past year with generative AI. We’re told it’s here so we just need to adjust to it, instead of admitting we do have the collective power to choose how it’s used — if it should be at all.

Some of the tech industry’s failures were because the products sucked. Juicero was an expensive plastic box that squeezed a bag of juice and Theranos’ finger-prick blood tests were a big fraud that eventually sent its founder to prison for 11 years. But crypto’s failure was a mixture of bad product and public opposition, while the metaverse was even more a rejection of a future where we would all be wearing headsets on our faces and socializing in virtual environments instead of with real people in meatspace. Like with Google Glass back in 2014, it was the public revulsion and open criticism that helped defeat those visions tech had for our collective future.

This was put on display recently with the Humane Ai Pin, a dumb aluminium square that you stick on the front of your shirt, use to projects menus on your palm, and is, of course, powered by AI. The product was roundly slammed by reviewers when they finally put it through some real tests earlier this month, but that was after months of reporting that gave the product the benefit of the doubt and echoed its marketing line that it might ultimately be a replacement for the smartphone. Oddly enough, the tech bro backlash to the criticism centered on YouTuber Marques Brownlee, or MKBHD.

Brownlee called the Pin the “worst product I’ve ever reviewed… for now.” He did a deep dive on how poorly the product worked and explained that viewers shouldn’t buy into a product simply because of the vision of what it hopes to be in the future. He’s absolutely right about that. Former AWS engineer Daniel Vassallo was not happy about it though, writing on Twitter that he found it “almost unethical” because reviewers have a duty to “do no harm” — which absolutely is not the case. Reviewers have a duty to be honest with their viewers or readers, though that’s often tempered by a desire to maintain a good relationship with major tech firms.

Kara Swisher’s Reality Distortion Field

In “Burn Book,” the longtime tech journalist tries to rewrite her story for the post-techlash era

DisconnectParis Marx

It’s funny that Brownlee was singled out in this case because he’s the kind of reviewer who will only slam a product if everyone else is going to do it too. He doesn’t often go hard on tech products and tends to give companies the benefit of the doubt, even when they don’t deserve it.

On the Vision Pro, for instance, he weighed the pros and cons while spinning a narrative that the product was indeed the future and was going to take off just like other Apple products before it. Meanwhile, the guy literally drives a Cybertruck and spends part of his review of the monstrosity explaining away the problems with Tesla’s build quality and criticizing those who point to videos of Cybertrucks stuck on a beach, struggling with going off road, and crashing into hotel signs in Beverly Hills for conflating driver error with vehicle deficiencies. The warning against buying into future visions expressed in the Humane Ai Pin review was nowhere to be found; he was all in on whatever future Apple or Tesla was ready to sell him.

No tech is inevitable

When Brownlee dismisses concerns with the Cybertruck or praises the future of what the Vision Pro could be, he’s not doing it because he’s being paid off by Apple or Tesla (though I think he certainly wants to keep his access to those companies). He does honestly believe what he’s saying, but his views are those of a tech booster. He wants to believe the narrative of tech inevitability, and he’s made a good career of echoing it to his followers — which is valuable to the tech companies he often features on his channel.

That narrative of inevitability is used to convince us we don’t have the power to challenge Silicon Valley and just need to accept whatever it throws at us. It might flow out of tech companies’ PR departments, but it’s echoed by well-connected journalists, influential tech publications, and even powerful political figures who effectively tell the public that resistance is futile and we must bow before our silicon overlords. But that’s clearly not true.

The Vision Pro and the Ai Pin are just the latest in a long line of tech failures, and they won’t be the last. We need to recognize that we have the collective power to challenge the industry, and no tech company or access journalist should be able to convince us otherwise. The only way we can hope to build a better future is by seizing that power and using it to rein in an industry that has already caused immense harm and is poised to do far more as it erodes the benefits of its products to squeeze out a bit more profit for greedy shareholders.

Vous connaissez OpenELM ? Non, normal, ça vient de sortir. Et c’est une famille de modèles IA open-source made in Apple conçus pour tourner directement sur vos appareils, sans passer par le cloud. En gros, c’est de l’IA maison dans nos iPhone, iPad et Mac…etc.

OpenELM combine plusieurs modèles de langage naturel (LLMs) utilisant des algorithmes évolutionnistes qui exploitent les principes techniques suivants :

1. Layer-wise scaling strategy : Cette stratégie consiste à allouer les paramètres dans les couches d’un modèle transformeur pour améliorer l’exactitude. Les modèles sont pré-alourés avec un budget de paramètres de 270 millions, 450 millions, 1,1 milliard et 3 milliards.
2. Pré-entraînement : Les modèles ont été pré-entraînés à l’aide d’une combinaison de datasets, incluant une sous-ensemble de Dolma v1.6, RefinedWeb, deduplicated PILE et une sous-ensemble de RedPajama. Ce dataset contient environ 1,8 trillion de tokens.
3. Evolutionary algorithms : Les algorithmes évolutionnistes sont utilisés pour combiner les modèles LLM et améliorer l’exactitude. Cela permet d’exploiter les forces combinées des modèles pré-alourés et d’améliorer leur précision.

Alors évidemment, Apple arrive un peu après la bataille dans l’IA, pendant que Microsoft et Google déboulent à fond la caisse. Mais bon, mieux vaut tard que jamais, et puis ils compensent avec du lourd, soit 8 modèles OpenELM au total, dont 4 pré-entraînés avec CoreNet et 4 fine-tunés. Et avec leur stratégie de scaling par couche ça optimise à fond l’allocation des paramètres.

Allez, je traduits… En gros, ça veut dire qu’ils sont hyper efficaces et précis. Prenez le modèle à 1 milliard de paramètres et bien bah il explose un modèle équivalent comme OLMo de 2,36% en précision, avec 2 fois moins de tokens en pré-entraînement. Et ce qui est top, c’est qu’Apple balance tout : code, logs d’entraînement, configuration…etc et pas juste le modèle final. Et vu qu’ils utilisent des datasets publics, c’est top en matière de transparence et vérification des biais.

En tout cas, une chose est sûre, avec OpenELM, Apple nous prouve qu’ils sont dans la course, et qu’ils comptent bien mettre le paquet sur l’IA

Et Merci à Letsar pour l’info, c’est lui qui m’a mis la puce à l’oreille sur OpenELM. Tu gères !

Source

Ça y est, les émulateurs de jeux rétro débarquent enfin sur l’App Store ! C’est la grande nouvelle annoncée par Apple ce vendredi. Les règles de l’App Store de la firme à la pomme s’assouplissent pour accueillir ces applications tant attendues par les fans de jeux rétro.

Jusqu’à présent, les émulateurs étaient persona non grata sur iOS, laissant les possesseurs d’iPhone dans l’obligation de passer par des solutions de contournement comme le jailbreak ou des magasins d’apps alternatifs. Mais ça, c’était avant ! Avec ce revirement, Apple ouvre grand les portes aux émulateurs déjà présents sur Android (enfin, ceux qui ont survécu aux assauts de Nintendo). Une décision qui pourrait bien couper l’herbe sous le pied des app stores tiers désormais autorisés dans l’Union Européenne.

Mais attention, pas question pour autant de faire n’importe quoi. Les jeux proposés devront respecter « toutes les lois en vigueur« , sous-entendu pas de place pour le piratage. Comme ça, les développeurs sont prévenus !

En parallèle, la firme de Cupertino a également clarifié sa position sur les « super apps » type WeChat. Mini-jeux et mini-apps devront se contenter donc se contenter d’HTML5 et n’auront pas le droit à leur app native.

Ces changements interviennent alors qu’Apple est dans le collimateur de la justice américaine. Le géant est accusé de vouloir étouffer la concurrence des apps de cloud gaming et des super apps. Une plainte qu’Apple semble avoir anticipée en autorisant récemment les services comme le Xbox Cloud Gaming ou GeForce Now.

Et ici, c’est la Commission Européenne qui met la pression. Résultat, Apple autorise désormais les apps de streaming musical à inclure des liens et des infos sur les abonnements hors App Store. Spotify peut dire merci à Bruxelles ! Mais tout n’est pas encore réglé. Malgré l’injonction de la commissaire européenne Margrethe Vestager de laisser les apps communiquer « librement » avec leurs utilisateurs, Apple compte bien continuer à prélever sa dîme sur les achats via les fameux liens. De quoi faire grincer des dents chez Spotify.

« Suivre la loi n’est pas optionnel, mais Apple continue de défier cette décision« , s’agace un porte-parole. « À partir du 6 avril, la Commission pourra lancer des procédures pour non-conformité et imposer des amendes quotidiennes. Il est temps d’agir de manière décisive pour donner une fois pour toutes un véritable choix aux consommateurs.« 

Bref, vous l’aurez compris, c’est un petit pas pour les émulateurs, un grand pas pour l’ouverture de l’App Store. Mais le chemin est encore long avant d’arriver à un véritable marché ouvert et concurrentiel sur iOS. En attendant, sortez vos manettes et préparez-vous à replonger dans la nostalgie du rétrogaming directement sur votre iPhone !

Source