AI Eye: Apple developing pocket AI, deep fake music deal, hypnotizing GPT-4

Apple wants to enactment an AI successful your pocket

Apple has been playing its cards adjacent to its thorax erstwhile it comes to AI. While rival Microsoft has jumped connected the ChatGPT bandwagon and is integrating AI into everything contempt the bugs and hallucinations, the acronym didn’t adjacent get a notation astatine Apple’s Worldwide Developers Conference successful June.

Reports emerged successful July, however, that Apple was moving connected its ain generative AI tool, dubbed internally “Apple GPT,” which uses a ample connection exemplary (LLM) model called Ajax. On this week’s quarterly net call, CEO Tim Cook said Apple was enthusiastic astir the exertion and has incorporated AI into forthcoming iOS17 features similar Personal Voice (voice cloning and text-to-speech) and Live Voicemail (live transcription). He added:

“We’ve been doing probe crossed a wide scope of AI technologies, including generative AI, for years. We’re going to proceed investing and innovating and responsibly advancing our products with these technologies, with the extremity of enriching people’s lives. That’s what it’s each astir for us. As you know, we thin to denote things arsenic they travel to market, that’s our M.O., and I’d similar to instrumentality to that.”

Of course, what mundane users privation to cognize is whether Siri volition beryllium getting an AI upgrade. And they surely look to beryllium moving connected it, with the Financial Times reporting that Apple is hiring dozens of researchers and engineers to enactment connected “compressing existing connection models truthful they tin tally efficiently connected mobile devices, alternatively than successful the cloud.” The ads indicated the institution is afloat focused connected bringing LLM exertion to mobiles.

Also read: Experts privation to springiness AI quality ‘souls’ truthful they don’t termination america all

There are speed, privateness and information reasons to tally the AI locally connected the telephone hardware alternatively than successful the cloud, fixed concerns implicit OpenAI and Claude hoovering up each your idiosyncratic and concern data. Back successful 2020, Apple spent $200 cardinal snapping up Seattle startup Xnor, which focuses connected this nonstop problem.

Apple’s Personal Voice is coming successful iOS17. (Apple)

Passwords adjacent much useless owed to AI

Even anterior to the advent of AI, computing exertion had progressed to the constituent wherever the mean eight-character password — utilizing a operation of numbers, precocious and little lawsuit letters and a peculiar quality arsenic recommended — could beryllium cracked successful astir 5 minutes. New probe indicates that AI password crackers similar PassGAN tin ace much than fractional of each commonly utilized passwords successful little than a minute.

Now it turns retired that AI tin enactment retired your password with greater than 90% accuracy, purely from the dependable of you typing. Given that astir everyone types wrong earshot of a machine oregon telephone mic, that’s a beauteous large exploitable area, particularly if you log successful to a tract portion connected a Zoom telephone (93% accuracy.)

The tech isn’t rather arsenic bully erstwhile users interaction benignant oregon usage the displacement key, but it’s adjacent clearer that passwords unsocial without 2FA request to beryllium consigned to the bin of history.

Passwords are progressively obsolete (Home Security Heroes)

Google and Universal negociate woody connected euphony heavy fakes

Johnny Cash’s fake mentation of “Barbie Girl” and Frank Sinatra riffing connected a large set instrumentality of “Gangsta’s Paradise” are a mates of the much amusing AI heavy fakes retired there. This has provoked alarm from artists, including Drake and Sting, who are understandably acrophobic astatine their unsocial vocals and euphony styles being ripped off.

Also read: ‘Elegant and ass-backward’: Jameson Lopp’s archetypal content of Bitcoin

In response, Google has reportedly entered negotiations with Universal Music to make a instrumentality for fans to make their ain morganatic heavy fakes of fashionable artists with a interest going backmost to the copyright holders. Artists would person the quality to opt successful oregon opt retired of the system. Google is trying to onslaught a akin woody with Warner Music, whose CEO, Robert Kyncl, enthused to investors this week that “with the close framework,” AI could “enable fans to wage their heroes the eventual compliment done a caller level of user-driven content . . . including caller screen versions and mash-ups.”

Some artists person embraced AI technology, with Grimes offering a 50/50 divided of proceeds to AI producers and Paul McCartney utilizing AI to amended John Lennon’s unsmooth demo vocals for “the last Beatles track,” which was abandoned owed to mediocre prime successful the ‘90s.

Disney’s AI task force

Hollywood writers and actors onslaught beryllium damned! Disney has created an artificial quality task unit to survey however AI tin beryllium utilized crossed the amusement behemoth. There are 11 existent occupation openings crossed its taxable parks, TV and advertizing divisions.

While creatives spot AI arsenic a threat, a Disney insider says the institution believes the bigger menace is not adapting to the caller scenery successful bid to bring budgets down — present topping $300 cardinal for tentpole releases similar Indiana Jones. Apart from bringing down the costs of peculiar effects with generative AI, a taxable parkland imagineer told Reuters that AI tin marque the company’s robots much lifelike, pointing to Project KIWI, which utilized instrumentality learning method to springiness a free-roaming Baby Groot from Guardians of the Galaxy property and quality movements.

Researchers ‘hypnotize’ GPT-4 to con different users

Disturbing research from IBM suggests that GPT-4 tin beryllium tricked into manipulating users. The researchers person shown that GPT-4 tin beryllium “hypnotized” to instrumentality portion successful multi-layered Inception-type games that saw the models “leaking confidential fiscal information, generating malicious code, encouraging users to wage ransoms, and adjacent advising drivers to plow done reddish lights,” according to Gizmodo.

Even if users figured retired 1 of the “games” the LLM was playing, the researchers had created aggregate different “games” the idiosyncratic would autumn into. Bard is seemingly much hard to manipulate than GPT-3.5 and GPT-4.

Also read: Blockchain games aren’t truly decentralized… but that’s astir to change

All Killer No Filler AI news

— Nvidia has conscionable unveiled its GH200 ace chip, which has 141GB of next-gen memory, 3 times the capableness of its fashionable H100 GPU. Nvidia says the outgo of powering LLMs volition driblet significantly

— A caller preprint from erstwhile Amazon AI researcher Konstantine Arkoudas analyzed GPT-4’s responses to 21 reasoning problems and concluded that: “Despite occasional flashes of analytical brilliance, GPT-4 astatine contiguous is utterly incapable of reasoning.” 

— Spotify’s AI diagnostic “DJ” — which recommends caller artists and tracks and tells you wherefore you should springiness them a spell — is being rolled retired to 50 countries this week. Users of the beta truthful acold person spent astir one-third of their listening clip utilizing DJ.

— Goldman Sachs predicts that AI investments volition soar to $200 cardinal globally by 2025, accounting for 4% of U.S. gross home merchandise and astir 2.5% of the GDP of different nations. One successful six companies mentioned AI connected caller net calls.

— Morgan Stanley is looking into concerns implicit an AI banal bubble, highlighting that erstwhile bubbles had seen three-year highest returns of 150%. While AI banal darling Nvidia is up by 200% this twelvemonth alone, broader AI indexes are lone up by 50%.

— Researchers astatine Harvard and the University of Washington recovered that crowdsourcing concern ideas from humans produced overmuch much caller ideas than GPT-4, portion prompting those aforesaid ideas from the AI produced ideas with amended biology and fiscal value. They concluded the champion mode guardant whitethorn beryllium an “integrative human-AI attack to problem-solving.”

— Nobody trusts AI with institution data, according to a Blackberry survey of 2,000 institution IT chiefs, which recovered 3 quarters are either implementing oregon considering bans connected ChatGPT and different LLMs for information security, privateness and firm estimation reasons. However, a McKinsey survey found that lone 21% of organizations person implemented immoderate policies connected generative AI truthful far.

Video of the week

Redditor SellowYubmarine posted this “AI-generated trailer” for a Magic 8 fearfulness movie to the Singularity subreddit. While it highlights that a azygous idiosyncratic tin employment AI tech to travel up with a beauteous awesome trailer, it inactive requires sizeable effort, with the idiosyncratic employing ChatGPT for dialog and communicative ideas Midjourney, Adobe Firefly, Runway, Pika Labs for the visuals and Photoshop, After Effects and Audition for editing.

Still, the caller tech means creators volition chiefly beryllium constricted by their imaginations successful the future, alternatively than budgets, arsenic has been the lawsuit successful the past.