Welcome to the newest edition of FindBiometrics’ AI update. Here’s the latest big news on the shifting landscape of AI and identity technology:
Meta’s AI chief, Yann LeCun, says LLM AI systems like OpenAI’s ChatGPT will never achieve Artificial General Intelligence due to their “very limited understanding of logic.” His Fundamental AI Research (FAIR) lab is taking an alternative “world modelling” approach to AI training to reach AGI, doing things like getting AI to predict what’s going to happen next in videos. LeCun’s position contrasts against recent news from OpenAI, which has seen a flurry of departures from AI safety researchers who are worried it’s moving too quickly toward AGI and even superintelligence.
Scale AI has raised $1 billion in its latest funding round, bringing its valuation to roughly $14 billion. The startup offers labelled data for AI training, and its work has attracted new investors Amazon, AMD Ventures, Intel Capital, and Meta. The latter also happens to be one of Scale’s biggest customers, alongside Microsoft and OpenAI.
Amazon is saving its money for a better chip. The tech giant was earlier planning to buy Nvidia’s Grace Hopper “superchip” for AI training, but has decided to hold off and wait for Nvidia’s recently announced Blackwell chip, which is expected to be twice as powerful. Morgan Stanley analysts are worried about “a pause in front of Blackwell” with respect to AI investment, but Nvidia has barely been able to keep up with demand as it is.
Coreweave is raising $7.5 billion in debt in a funding round led by Blackstone and Magnetar Capital. The New Jersey-based firm offers AI computing services, and is working to double its number of data centers to 28 by the end of this year. Co-founder and CEO Michael Intrator said the “calibre of investors” in the debt financing round speaks to “the insatiable market appetite for AI infrastructure.”
Huawei has signed up over a dozen startups through its “AI-in-a-box” business line, which offer tech bundles allowing AI companies to run their AI systems on premises in private cloud environments, negating the need to hire expensive cloud computing services. It’s an approach that is particularly popular in China, where the so-called “all-in-one machines” market is expected to reach $2.3 billion this year.
Taiwan’s science and technology minister, Wu Cheng-wen, is insisting that Taiwan Semiconductor Manufacturing Company (TSMC) will retain its core R&D work in the country even as it seeks to build out manufacturing infrastructure internationally, in places like the US, Canada, France, and Germany. TSMC is hugely important in AI infrastructure, but partners are wary about geopolitical risk stemming from China. CEO CC Wei recently said customers would need to pay a premium to have their tech made “in a certain geographical area”.
Scarlett Johansson got OpenAI to take down one of its flagship new voice assistant models, suggesting (through legal counsel) that it was designed to sound like her. “In a time when we are all grappling with deepfakes and the protection of our own likeness, our own work, our own identities, I believe these are questions that deserve absolute clarity,” the actress said in a statement.
The European Union’s AI Act will come into force next month, having received final endorsement from EU countries this week. Among other things, the Act will establish robust transparency requirements for “high-risk” AI systems and restrict the real-time use of facial recognition by law enforcement agencies, with such application allowed only in certain scenarios.
The chatbot’s take: We decided to test out GPT-4o’s logic skills, and got the right answer along with some social commentary.
–
May 22, 2024 – by Alex Perala
Follow Us