Artificial Intelligence - Page 34

AMD and Fujitsu have just announced a new strategic partnership to create computing platforms for AI and high-performance computing (HPC).

The new partnership includes aspects from technology development to commercialization, and will further boost the creation of open source and energy-efficient platforms using advanced processors with "superior power performance and highly flexible AI/HPC software" that aims to accelerate open-source AI and HPC initiatives.

The insane rise of AI including generative AI, cloud service providers (CSPs) and users are wanting optimized architectures, at various price and power-per-performance configurations explains AMD in the press release.

Apologies have already been issued by the people behind the event that misled thousands of Dubliners out of their homes to a fraudulent Halloween parade.

The Halloween parade was organized and listed on the My Spirit Halloween website, where it stated the Mácnas parade was occurring between 7pm and 9pm on Thursday last week. Thousands of people showed up to the event and reports by people on the ground, including video, show an incredibly busy street with many rows of people deep that are blocking traffic from using the road. Participants of the event were then informed by the Irish police that no event was happening and were ordered to disperse.

It now appears a member of the team organizing the event simply cut and pasted the notice for last year's event onto the calendar for 2024. Additionally, a Pakistan-based company has issued an apology to Dubliners for the phony event, claiming the mishap occurred due to "human error" on the website. A Pakistan-based man behind the website apologized to the misled people, saying he and his team is "highly depressed", "embarrassed," and "very sorry".

SK hynix has just unveiled the world's first 16-Hi HBM3E memory, which will arrive in up to 48GB capacities per stack.

The new 16-Hi HBM3E memory was announced by SK hynix CEO Kwak Noh-Jung during the SK AI Summit 2024 event, with the SK hynix CEO unveiling the 16-Hi HBM3E memory first with samples of 48GB capacity. This is the highest capacity, and the highest number of layers ever in the HBM memory industry.

This comes in hot off the heels of a story from yesterday, in which NVIDIA CEO Jensen Huang reportedly called SK hynix leadership and told them to bring up supply of its next-gen HBM4 memory by 6 months. SK hynix's latest 16-Hi HBM3E memory is "expected to open up from the HBM4 generation" with the company developing 48GB 16-Hi HBM4E memory in a "bid to secure technological stability" with plans to provide samples to customers in early 2025.

NVIDIA CEO Jensen Huang called up SK hynix chairman Tae-won to bring forward its supply of next-generation HBM4 memory by 6 months.

Originally, SK hynix planned to have its next-gen HBM4 memory ready in the second half of 2025, but now NVIDIA is pushing the South Korean memory giant to make its HBM4 now, now, now. NVIDIA currently uses SK hynix's bleeding edge HBM3E memory on its new Blackwell B200 and GB200 AI chips, with HBM4 memory planned to be inside of its next-gen Rubin R100 AI GPU.

SK hynix had its next-gen HBM4 reportedly ready to tape out in October, something we reported on in August, and it looks like a call from Jensen will be enough to kick HBM4 into reality from SK hynix earlier than planned... 6 months ahead of time is a big ask from NVIDIA, but they know SK hynix is at the top of its game and ready for HBM4.

Mark Zuckerberg has provided a small update on Meta's work on its new Llama 4 model, which is being trained on a cluster of AI GPUs "bigger than anything" Zuck has seen.

Meta is cooking its new Llama 4 right now, with Zuckerberg telling investors and analysts on an earnings call this week that the initial launch of Llama 4 is expected later this year. Zuck said: "We're training the Llama 4 models on a cluster that is bigger than 100,000 H100s, or bigger than anything that I've seen reported for what others are doing. I expect that the smaller Llama 4 models will be ready first".

Meta's new AI supercomputer with its 100,000+ NVIDIA H100 AI GPUs reportedly cost over $2 billion for the H100 AI GPU chips alone, which means Mark Zuckerberg is signing some fat cheques to NVIDIA. Speaking of which, NVIDIA CEO Jensen Huang recently spoke with Meta CEO Mark Zuckerberg, where Jensen said Meta now has 600,000+ NVIDIA H100 AI GPUs to which Zuck replied saying that Meta were "good customers for NVIDIA".

Microsoft has delayed its Recall feature yet again, amid all the controversy, the company pushed its big Copilot+ PC feature into December (as it's now missed its promised October release)

Brandon LeBlanc, senior product manager of Windows, in a statement to The Verge: "We are committed to delivering a secure and trusted experience with Recall. To ensure we deliver on these important updates, we're taking additional time to refine the experience before previewing it with Windows Insiders".

The company originally pulled its Recall feature from launch over security concerns -- it was logging everything, like, everything, everything -- ahead of the Copilot+ PC launch with Qualcomm. Now there are AMD and Intel processors with NPUs inside of laptops, and the big Recall feature launching in October, is delayed into December.

Elon Musk's xAI startup is currently in the process of upgrading its Colossus AI supercomputer cluster from 100,000 NVIDIA Hopper AI GPUs, doubling it to an insane 200,000 NVIDIA Hopper AI GPUs.

Colossus is the world's largest AI supercomputer, and is using used to train xAI's Grok family of LLMs (large language models), with chatbots on offer for X Premium subscribers. Elon's massive xAI Colossus supercomputer cluster facility was recently toured (more on that in the links below) and took just 122 days to complete, something NVIDIA CEO Jensen Huang recently called Elon Musk "superhuman" because of it.

NVIDIA recently posted some content explaining its partnership with Elon and xAI, with the company explaining: "The supporting facility and state-of-the-art supercomputer was built by xAI and NVIDIA in just 122 days, instead of the typical timeframe for systems of this size that can take many months to years. It took 19 days from the time the first rack rolled onto the floor until training began".

In a recent episode of the "On With Kara Swisher" podcast, Oscar winner and Marvel superstar Robert Downey Jr. revealed he plans to sue any Hollywood executive who uses his likeness in the form of a digital AI replica to recreate him after his death.

Downey was asked about what he thinks the future of Hollywood will look like with the emergence of AI technologies and how he feels about the technology being used to recreate popular actors after they have passed away. The topic presented itself through questions asked about Downey and his tenure as Iron Man. Downey said he is confident that Marvel executives won't recreate him as Iron Man, as he believes the people in control know him very well and "they would never do that to me, with or without me."

However, when asked 'what about future executives?' Downey acknowledged there will certainly be a push, at one stage or another, for his likeness to return to the Marvel Cinematic Universe (MCU) after he has passed away. But that push by future executives will be met with legal ramifications by Downey's law firm, as the Oscar-winning actor explained,"I would like to here state that I intend to sue all future executives just on spec."You'll be dead," Swisher noted. "But my law firm will still be very active," responded Downey

OpenAI is working with Broadcom and TSMC on building its first in-house chip designed to support its extensive AI systems.

The news is coming directly from Reuters, which is reporting from its sources who "requested anonymity as they were not authorized to discuss private matters" (but I guess, did so with one of the largest publications in the world). OpenAI did consider building everything in-house, with CEO Sam Altman having ambitious plans to have a global network of dedicated chip fabrication plants, but that is now not happening.

Reuters reports that OpenAI "dropped the ambitious foundry plans for now due to the costs and time needed to build a network, and plans instead to focus on in-house chip design efforts".

Linux creator Linus Torvalds has come out blasting AMD, Intel, and NVIDIA recently calling their hardware "buggy" and now has said that the AI industry is full of it and that it's "90% marketing" and "10% reality".

During a recent interview at the Open Source Summit with TFIR, Torvalds talked about the hype surrounding AI, and how the technology has been shadowed by the overwhelming marketing of tech giants (of which I agree).

He said: "I think AI is really interesting and I think it is going to change the world. At the same time, I hate the hype cycle so much that I really don't want to go there. So, my approach to AI right now is I will basically ignore it because I think the whole tech industry around AI is in a very bad position, and its 90% marketing and 10% reality. And, in 5 years, things will change and at that point, we will see what of the AI is getting used for real workloads".

Elon Musk's gigantic Colossus AI supercomputer from his xAI startup is powered by 100,000 x NVIDIA H100 AI GPUs, and has just had an awesome walkthrough by our friends at ServeTheHome. Check it out:

Patrick from ServeTheHome notes that the engineering accomplishment from Elon and his team at xAI is "absolutely amazing". xAI only took 122 days to build out the insane 100,000 NVIDIA H100 AI GPU-powered Colossus AI supercomputer, something that normally takes many years... but not for Elon and xAI, which is why NVIDIA CEO Jensen Huang recently called the SpaceX and Tesla boss "superhuman". More on that below:

Back to the xAI supercluster, with STH noting that the basic building block for Colossus is the Supermicro liquid-cooled rack, which features 8 x 4U servers each with 8 x NVIDIA H100 AI GPUs for a total of 64 x H100 AI GPUs per rack. 8 of these GPU servers + Supermicro Coolant Distribution Unit (CDU) and required hardware make up one of these GPU compute racks.

According to a new report, Meta has been indexing the web for at least eight months. The company aims to have its own search engine that can be called on and integrated into Meta AI. This will give its chatbot and other generative AI tools an alternative to Google Search and Microsoft Bing while decreasing the company's reliance on the services of its competitors.

Meta has been using web crawlers for a while, which are part of Facebook and 'crawl' the content and information of links shared on Meta's social media platform. This is different in that it would be used primarily for AI search, and as such, it would need to scrape and crawl the entire internet for information and training.

The company has yet to detail its search engine plans formally. Still, it recently announced a multi-year partnership with a news outlet, Reuters, for the Meta AI chatbot to use the source with citations when answering questions.

Google will be updating its Gemini AI model with a more powerful version later this year, which is to be expected. However, according to a new report (via The Information and Reuters), the update will include 'Project Jarvis'. It's not the AI butler that serves Tony Stark in the Marvel Cinematic Universe, but it's close.

According to the report, Project Jarvis is a new AI agent set to become part of Google's Chrome web browser. It can browse various websites for you, summarize the content, fill out web forms, and even make purchases. Now, if it can do all that, you'd have to assume that it could solve those "I'm not a robot" tests where you're tasked with finding buses and traffic lights in the internet-age equivalent of Where's Waldo.

Naturally, if it can go online and buy you a new pair of socks or an OLED TV, you can fine-tune its behavior and set restrictions.

NVIDIA CEO Jensen Huang joined the King of Denmark to launch the country's largest sovereign AI supercomputer, which is aimed at breakthroughs in quantum computing, clean energy, biotechnology and other areas serving Danish society and the world.

Denmark's first AI supercomputer has been dubbed "Gefion" after a goddess in Danish mythology, as an NVIDIA DGX SuperPOD powered by 1528 NVIDIA H100 Tensor Core AI CPUs, interconnected using NVIDIA Quantum-2 InfiniBand networking.

Jensen said: "Gefion is going to be a factory of intelligence. And this factory of intelligence is a new industry that never existed before; it sits on top of the IT industry - we're inventing something fundamentally new. Denmark recognizes that to innovate in AI, the most impactful technology of our time, it must foster a domestic AI infrastructure and ecosystem. The Gefion supercomputer will supercharge the scientists of Denmark with local AI computing infrastructure to drive advancements in life sciences, climate research, and quantum computing".

NVIDIA will be supplying its new Blackwell AI GPUs to Indian companies including Mukesh Ambani's Reliance Industries, to build a new AI supercomputer in India.

Reliance is building a new 1 GWh (one-gigawatt hour) data center in the western state of Gujarat, India, announced at an AI summit recently held in the business capital of Mumbai. NVIDIA CEO Jensen Huang and Mukesh Ambani announced the news, with Jensen saying: "In the future, India is going to be the country that will export AI. You have the fundamental ingredients - AI, data and AI infrastructure, and you have a large population of users".

Jensen added: "India is already world-class in designing chips, India already develops AI. Instead of being an outsourcer and a back office, India will become an exporter of AI".

NVIDIA CEO Jensen Huang says that AI will do some jobs with 1000x higher productivity than humans, but AI will never fully replace the humans that perform these jobs.

At NVIDIA's October AI Summit held in Mumbai, India, CEO Jensen Huang said: "As we speak, AI has no possibility of doing what we do. Depending on the jobs we do, it could do 20% of our jobs 1000 times better. For some people, it could do 50% of their job 1000x better. But in no job can they do all of it".

Jensen was asked if AI would take his job -- as the CEO of NVIDIA -- to which he replied: "absolutely not".

OpenAI has suffered yet another blow, as a senior staffer in the company's AGI Readiness team, a team dedicated to advising OpenAI on the impact of the powerful AI models it's creating and how ready the world is for them, has left the company. This was promptly followed by a warning published to the former OpenAI's staffer's Substack account.

The former OpenAI senior staffer is Miles Brundage, who, as of Friday this week, will no longer be working at OpenAI's AGI Readiness team. For those that don't know, AGI stands for Artificial General Intelligence (AGI), which is the description of an AI model with the same level of cognitive abilities as a human across all fields. This level of sophistication has yet to be fully achieved, but given the potential impact of such a system coming online or potentially falling into the wrong hands, guardrail teams such as the AGI Readiness team were formed.

However, Brundage states in his post that OpenAI has "gaps" in its readiness policy, but they aren't alone in this problem as every other AI lab also does. According to Brundage, OpenAI, and any other AI company, along with the world, isn't ready for AGI. Additionally, the post by the former OpenAI staffer revealed his departure triggered a complete disbanding of the AGI Readiness team, which comes at a time when OpenAI is attempting an internal restructuring into a for-profit business.

Microsoft has taken to its blog to announce the release of a suite of autonomous artificial intelligence agents that will serve businesses as virtual employees.

Redmond states in its blog post that it's announcing new capabilities that will enable customers to create autonomous agents with Copilot Studio, along with ten new autonomous agents in Dynamic 365. Microsoft writes that agents should be thought of as the "new apps for an AI-powered world" and that it believes one day, every organization will have a "constellation of agents" that will range from simply prompt and response bots to bots that are completely autonomous.

Copilot will be how customers interact with these agents, and their capabilities will range from sales, supplier communications, customer intent, and customer knowledge management agents. Microsoft states in its blog post that its AI agents will be able to increase the productivity of a business and is an example of how artificial intelligence can increase the output generated by a worker per hour. As for custom agents, Microsoft chief executive Satya Nadella said that Copilot Studio is built as a "no-code way for you to be able to build agents," which means users won't need any prior programming knowledge to create a custom agent successfully.

The Federal Trade Commission (FTC), the US government agency designed to protect American consumers against illegal business practices, has changed its guidelines to target AI-generated content.

Since the rise of AI-generation tools such as OpenAI's ChatGPT, Microsoft's Copilot, or simple image-generation tools such as DALLE, internet users have been enduring the rise of AI-generated content. While this is typically fine, as most of the content is interesting images or videos on social media timelines, there is a real problem when it comes to posting a review for a product or service when AI created it. The FTC's new guidelines are designed to protect consumers against businesses intentionally trying to mislead consumers.

An example of this would be a business that uses AI tools to create fake reviews or testimonials for products. These AI-generated reviews would then be published all over the listing for the product and would mislead potential buyers into thinking the product is more reliable than it actually might be. More specifically, the new FTC guidelines list fake reviews attributed to people who don't exist or someone that overstates their level of experience with the product as a particularly egregious offence.

The HBM memory business is expected to continue to explode throughout 2025, with experts predicting the HBM market will grow 156% in 2025 to reach $46.7 billion... a market worth just $18.2 billion in 2024.

At the TrendForce Roadshow Korea held in Seoul last week, senior vice president of research operations at TrendForce, Avril Wu, said she expects to see the global HBM memory chip business to expand by 156% next year to $46.7 billion, up from $18.2 billion this year. HBM memory chip share of the overall DRAM market is expected to rise up to 34% in 2025, up from 20% in 2024.

NVIDIA's new Blackwell AI chips are the driving force behind ultra-fast HBM memory, with TrendForce noting that major AI solution providers will witness a "significant shift" in HBM specification requirements towards HBM3E, with the boost to HBM3E 12-Hi stack products, and will increase the HBM capacity per chip.