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 has nearly completed the construction of its own hyperscale AI cluster, which will reportedly exceed 100,000 Nvidia H100 GPUs. The news comes after Elon Musk raised eyebrows last week with his announcement that his AI startup, xAI, had established such a cluster, and that he expected it to double in size in a few months. Mark Zuckerberg’s team plans to train the next version of its Llama AI model on its supercomputer cluster.
OpenAI has 11 million paying subscribers, according to COO Brad Lightcap. That’s 10 million regular paying subscribers and another million who are subscribed to more expensive enterprise plans. The company is in talks to raise as much as $7 billion in a new funding round that could feature MGX, a fund backed by the United Arab Emirates.
It also launched a new “o1” model of its flagship GPT AI system that is optimized for reasoning tasks. OpenAI says o1 scored 83 percent on a qualifying exam for the International Mathematics Olympiad, compared to GPT-4o’s 13 percent. Remarkably, CTO Mira Murati said o1 offers “visibility into the model’s thinking… we can observe its thought process, step by step.” It’s helping the OpenAI team understand how its product works.
China-based chipmakers Enflame and Biren are planning IPOs on Shanghai’s STAR Market stock exchange, aiming to go public this year or in early 2025. Their AI chips are generally considered inferior to those of Nvidia, but are presumably benefitting from a government push to support the domestic tech industry, and Enflame has received significant investment from Tencent, while Biren’s investors include Ping An Insurance Group Co. and IDG Capital.
TSMC saw a 33 percent increase in revenues last month, and is expected to see 37 percent growth in Q3. The Taiwanese chipmaker continues to benefit from AI excitement. It’s the primary manufacturer of the iPhone’s main processor, and Apple has been betting big on integrating AI into its newest device models.
Sixty countries including the United States have endorsed an international “blueprint” for AI safety in military applications. The document was discussed at the Responsible AI in the Military Domain (REAIM) summit in Seoul, which followed an early discussion in The Hague last year that issued a “call to action” for countries to agree on things like risk assessments and human control. China had signed on for that one, but declined to back the new blueprint.
A number of AI companies have agreed to a White House call for “voluntary commitments from AI model developers and data providers to reduce AI-generated image-based sexual abuse.” Signatories include Adobe, Cohere, Microsoft, Anthropic, and OpenAI, who committed to things like integrating “feedback loops” into AI training to prevent the generation of deepfake porn. It’s a self-policing commitment, and some notable image generation specialists like Midjourney and Stability AI declined to sign on.
The chatbot’s take: This week, we decided to test out the reasoning capabilities of OpenAI’s new model:
–
September 13, 2024 – by Alex Perala
Follow Us