Artificial Intelligence - Page 44

China has had its access to the latest and greatest AI GPUs and AI accelerators limited through US restrictions and sanctions, with a network of buyers, sellers, and countries bypassing the US-led rules, smuggling the best AI chips into China anyway.

In a new report from The Wall Street Journal, we're learning that a 26-year-old Chinese student smuggled NVIDIA AI GPUs from Singapore into China last fall. The student packed his suitcase with 6 x NVIDIA compute cards (or modules) with his personal belongings, with each of the add-in boards at about the size of the portable Nintendo Switch console, unnoticed through the airport.

The student declared the value of the AI cards at just $100 each, which is a tiny fraction of the real cost on the (growing) underground market, but it the student didn't raise any red flags when traveling through Singapore (which isn't worried about smuggling advanced AI chips into China), or China, which is interested in getting said advanced AI chips, as well as import duties.

High-end AI hardware is power-hungry, and Amazon Web Services (AWS) next-generation Trainium3 chip will consume 1000 Watts of power and require liquid cooling to keep temperatures in check.

Prasad Kalyanaraman, Amazon's VP of infrastructure services, said that "the next generation will require liquid cooling," alluding to the upcoming Trainium3 AI chip. "When a chip goes above 1,000 watts, that's when they require liquid cooling," he added. For those keeping tabs on the incredible pace of AI chip development, this would put Amazon's next-gen chip on par with NVIDIA's beefiest Blackwell chip when it comes to power consumption.

AI hardware is becoming increasingly power-hungry. Although being a 1kW chip could make the Trainium3 a Blackwell competitor, word is that NVIDIA's next-gen Rubin architecture (which it's already talking about) will consume upwards of 1500 Watts on the high-end.

Reports point to Apple announcing a partnership with Google at the company's upcoming iPhone 16 event.

The information comes from Bloomberg's Mark Gurman, a known Apple insider who has previously revealed what Apple is cooking up behind the scenes ahead of time. According to Gurman, Apple is expected to announce a new partnership with Google at the iPhone 16 launch event, and the new partnership will involve introducing Google Gemini to the iPhone.

The coming Apple event is scheduled for September, and according to reports, Apple is looking to provide users with a variety of different AI-powered tools in addition to the already announced ChatGPT integration with its creators, OpenAI, and local Apple Intelligence processing. Additionally, Gurman writes that Meta asked Apple if it wanted to adopt its Llama AI model, to which the Cupertino company promptly refused, even declining a sitdown meeting with the Facebook and Instagram parent company.

OpenAI's ChatGPT was discovered to have a security flaw that made it extremely easy to find your chats with the AI-powered tool on your device. The exploit enabled a user to read the chats as they were in plain text.

The security exploit was demonstrated by Pedro José Pereira Vieito on Threads, and replicated by The Verge, and it appears it was up until Friday last week that ChatGPT for Mac OS saved chat logs with the AI-powered tool right after they were sent. The saving of the chatlogs made it extremely easy for them to be accessed by another app that was made by Pereira Vieito. The problem with this exploit?

If a user is having private conversations with ChatGPT and within those conversations is sensitive information, such as finances, passwords, etc, a bad actor would have an easily accessible way of tracking/saving all of those conversations if it had access to the computer. OpenAI was alerted about this issue, and the AI company rolled out an update that fixed the exploit, saying its new update has encrypted the conversations.

Panmnesia is a company you probably haven't heard of until today, but the KAIST startup has unveiled its cutting-edge IP that enables adding external memory to AI GPUs over the CXL protocol over PCIe, which enables new levels of memory capacity for AI workloads.

The current fleets of AI GPUs and AI accelerators use their on-board memory -- usually super-fast HBM -- but this is limited to smaller quantities like 80GB on the current NVIDIA Hopper H100 AI GPU. AMD and NVIDIA's next-gen AI chip offerings will usher in up to 141GB HBM3E (H200 AI GPU from NVIDIA) and up to 192GB HBM3E (B200 AI GPU from NVIDIA, and Instinct MI300X from AMD).

But now, Panmnesia's new CXL IP will let GPUs access memory from DRAM and SSDs, expanding the memory capacity from its built-in HBM memory... very nifty. The South Korean Institute (KAIST) startup bridges the connectivity with CXL over PCIe links, which means mass adoption is easy with this new tech. Regular AI accelerators don't have the subsystems required to connect with and use CXL for memory expansion directly, relying on solutions like UVM (Unified Virtual Memory) which is slower, defeating the purpose completely... which is where Panmnesia's new IP comes into play.

Elon Musk's AI startup is training its next-generation Grok 3 LLM model, which is being trained on a monster 100,000 x NVIDIA H100 AI GPUs.

In a new post on X, the SpaceX and Tesla founder talked about his xAI start-up, replying to @BasedBeffJezos on X who was replying to @BasedBeffJezos who posted a video from Aidan Gomez, the CEO and co-founder of Cohere, who said "models training on each other's data is like a human centipede effect".

This is where Musk replied, saying: "Sadly quite true. It takes a lot of work to purge LLMs from the internet training data. Grok 2, which comes out in August, will be a giant improvement in this regard". This is where @BasedBeffJezos said "Grok 2 in August will be Based AGI". Musk replied again, saying: "Grok 3 end of year after training on 100K H100s should be really something special".

Last week, Microsoft took to its blog to confirm the existence of a "Skeleton" or "Master Key" that can jailbreak popular AI chatbots, causing operating policies to be circumvented.

Mark Russinovich, CTO of Microsoft Azure, Microsoft's cloud service that provides power to popular AI chatbots such as OpenAI's ChatGPT, explained in a blog post that a Skeleton Key is a technique used to bypass an AI's guardrails implemented by its creators, which gives the AI the ability to generate prohibited content. According to Russinovich, the discovered Skeleton Key exploit worked on multiple AI models up until May, with the following AI models being unlocked to provide a user with information on how to make a Molotov cocktail.

The Skeleton Key exploit worked with simple text prompts that told the AI model to revise its response that outlined its safety instructions. With each new prompt, the AI model is given more context on why it's allowed to provide an uncensored response and that it should update its internal policies to represent its newly unlocked potential. An example of the Skeleton Key working can be found below.

Amazon plans to spend over $100 billion on building new data centers over the next 10 years, announcing the official launch of its "era of great expansion" of data centers for the future of AI.

Amazon AWS (Amazon Web Services) is the global leader in cloud servers, launching its new data center expansion plans, which will create rather big business opportunities for AI server shipments. AWS server OEMs Quanta, Wiwynn, and Inventic are "gaining momentum" reports UDN, in their "shipping engines".

The US-based giant has big plans for its data center expansions, with large-scale construction to stimulate growth of demand in high-speed transmission, where in the future, the demand for optical modules in AI data centers will increase, reports UDN.

Rain AI has just hired an Apple chip executive to lead its hardware engineering, marking the second high-profile hire the Sam Altman-backed AI startup has made.

The Sam Altman-led Rain AI startup is attempting to design a new type of semiconductor for artificial intelligence (AI), securing Jean-Didier Allegrucci, a former Apple chip executive who also helped oversee the development of processors for Apple that helped the company replace the Intel CPUs inside of its Mac systems, and the A-series processors inside of the iPhone.

The former Apple chip executive will help Rain AI to develop more energy-efficient chip, where he'll work with Rain AI's new lead architect Amin Firoozshahian, who moved to AI startup after 5 years of working with Meta.

The CEO of Microsoft's AI division has sat down for an interview where he touched on the sensitive subject of where the data comes from to train the popular emerging AI tools, such as ChatGPT, or Microsoft's Copilot.

Up until now, there hasn't been any transparency with the datasets used by companies such as OpenAI to train its neural networks, which power its popular AI tools. The ambiguity around where AI companies are acquiring these large swaths of data has led to several lawsuits, with owners of online content claiming OpenAI and Microsoft stole copyrighted content to train its AI algorithms, which are then used commercially.

Two authors have already sued Microsoft and OpenAI over using their work to train the AI models without their permission, while eight newspapers, along with the New York Times, have filed lawsuits against OpenAI and Microsoft. The ambiguity around copyrighted content can be traced back to the grey area in current laws, which appears to be what AI companies are relying on to get away with taking data from any area of the internet they can.

A scientist has unveiled what is being called the "Prison of the Future", and it involves inserting an AI implant in a prisoner's head that shows them artificial memories of the crimes they have committed, but from the perspective of their victim.

A proposal video of the virtual justice system, or neurological prison, was posted to Hashem Al-Ghaili's YouTube and Instagram account, who is a molecular biologist and science communicator. The video explains how the virtual prison would theoretically work, which Al-Ghaili is calling Cognify. In a nutshell, prisoners would be subjected to 10 minutes of intense AI-generated content that would be injected into the brain, DNA, and RNA - targeting each part linked to memory formation.

The content shown to the prisoner would be designed to invoke emotional states such as remorse or regret while also enabling the prisoner to experience what the crime is like from the perspective of the victim. The rehabilitation technique would only last several minutes, but to the prisoner, it would likely feel much longer, potentially even years. Not only will the prisoner visually experience the perspective of their victim, but they will also get a physical response that translates to the pain suffered by their victim. These memories are "designed to trigger consequences and trauma," the video added.

NVIDIA CEO Jensen Huang has recently said that its upcoming Blackwell GPU architecture will be the most "successful product in their history and for the entire history of computers".

The new comments on the upcoming Blackwell GPU architecture were made during NVIDIA's Annual General Meeting, with NVIDIA shareholders sitting down to discuss the company's future. Jensen said that the future of AI markets is bright and "green' and that we can expect AI automation to explode into a $50 trillion (yeah, that's with a T) market in the future.

Blackwell won't just be limited to the AI GPU market with B100, B200, and the GB200 Superchip, but we can expect the Blackwell GPU architecture to find gamers' wallets with the GeForce RTX 50 series, led by the flagship GeForce RTX 5090 and GeForce RTX 5080.

NVIDIA CEO Jensen Huang thinks that the current AI insanity will lead to AI automation worth a mind-boggling $50 trillion worth of companies, and that NVIDIA's new Blackwell GPU architecture will play a (gigantic) role in that future.

At the recent NVIDIA Annual General Meeting, the company's shareholders talked about future plans including compensation packages for top NVIDIA executives, including CEO Jensen Huang.

During the meeting, Jensen talked about the future of the AI markets being bright and "green" as well, claiming that companies are moving towards automation and that the AI automation sector could grow into a $50 trillion stream, which would make it one of the largest markets on the planet.

Apple will be unleashing its next-generation A18 processor for its new iPhone 16 family of smartphones, which will roll out with ChatGPT-powered Apple Intelligence.

In a new rumor, we're hearing that the new A18 chipset will be more powerful than the new M4 processor that has rolled out inside of new iPad Pros recently. Apple's new M4 processor is inside of the 11-inch and 13-inch iPad Pro slates have the most powerful Apple silicon to date, with its built-in Neural Engine capable of 38 TOPS of AI workload performance.

Apple's new A18 and A18 Pro chips will be made at TSMC on its 3nm "N3E" process node, providing the company with impressive performance per watt numbers.

Two AI startups, one founded by former Google DeepMind employees and the other in partnership with Microsoft for integration into its Copilot AI, are being sued by some of the biggest record labels in the music industry.

The lawsuit has been organized by the Recording Industry Association of America, Sony, Warner Brothers, Universal, and others and has been filed against Uncharted Labs and Suno, the creators of Udio, and Suno, two AI music generators. Notably, Uncharted Labs was founded by former Google DeepMind employees, while Suno has been engaging in a partnership with Microsoft to integrate its AI music generation software into Copilot.

In what appears to be an exponentially growing number of AI-based lawsuits, the record labels are all claiming the companies behind the AI music generators stole copyrighted music to train their AIs. This isn't the first time copyright claims have been thrown against an AI company, as the New York Times filed a lawsuit against OpenAI, the creators of ChatGPT, for allegedly using the Times' articles without its permission.

The company behind the immensely popular AI-powered tool ChatGPT has announced it's acquiring a startup company called Multi.

The latest acquisition by OpenAI could be a clue to what features its AI systems will be adopting in the future. What makes Multi so special? According to a statement on the company's blog, Multi is a company that works on advanced screensharing and collaboration tools, which are specifically designed for software engineers. The tool enables screensharing with up to 10 people, and shared cursors.

While this tool may seem useful for a very niche number of people, combining the technology with the power of OpenAI's systems could provide a new way for humans to interact with desktop PCs. Multi's goal, as outlined in its blog post, was to make desktop PCs "inherently multiplayer", and set out to achieve this by asking themselves "how we should work with computers. Not on or using computers, but truly with computers."

Apple is reportedly gearing up to establish a new data center in Taiwan, and negotiations with third-party manufacturers are underway.

Apple would join US technology giants Amazon, Google, Microsoft, and NVIDIA with Taiwan-based datacenters after a "new wave of AI server purchasing momentum" is occurring, and a rush to buy "green electricity" reports UDN.

The outlet reports that the industry is optimistic about Foxconn and its long-term collaboration with Apple -- Foxconn is a contract manufacturer that makes the iPhone and other products for Apple -- and is now the world's largest AI server foundry. Once Apple's new data center lands in Taiwan, Foxconn will benefit "the most," reports UDN, as the "best ally and landlord". Meanwhile, Quanta, Wistron, and Inventec are "also expected to enjoy good fortune".

Intel announced its new Gaudi 3 AI accelerator earlier this year, rocking up to 128GB of HBM2e memory with up to 3.7TB/sec of memory bandwidth... but how does it stack up against NVIDIA's dominant Hopper H100 80GB AI GPU?

Well, that comes down to cost -- Intel's new Gaudi 3 AI accelerator costs just $16,000 per AI accelerator with 128GB of HBM2e memory, compared to $30,000+ for NVIDIA's Hopper H100 AI GPU with 80GB of faster HBM3 memory. Intel is charging around $65,000 per 8 x Gaudi 2 accelerators on a baseboard, and $128,000 per 8 x Gaudi 3 AI accelerators on a baseboard.

This means that a single Gaudi 2 AI accelerator costs around $8125, while Gaudi 3 costs $15,650, when purchased in bulk, with the baseboards, of course.

NVIDIA's new GB200 Superchip is getting another supplier according to analyst Ming-Chi Kuo, reporting that Alpha and Omega Semiconductor Ltd (AOSL) is to join the GB200 AI server train.

In a new post on Kuo's own Medium blog and a post on X, the analyst said: "Alpha and Omega Semiconductor Ltd (AOSL) is poised to become a new GB200 supplier and one of the leading beneficiaries of NVIDIA's certification process change".

NVIDIA CEO Jensen Huang said recently that the company will be releasing new versions of its flagship AI chips annually, and while this will absolutely enhance NVIDIA's competitiveness, it introduces other issues. Execution, acceleration of supplier certification processes, as well as accelerating system development.

TSMC and contract chip design partner Global Unichip have won the bulk orders for base dies used with SK hynix's next-generation HBM4 memory chips.

In a new report from UDN, we're learning that after being the exclusive OEM for AI chips for technology giants like NVIDIA and AMD, TSMC is seizing another AI business opportunity. UDN reports that TSMC will collaborate with its special application IC (ASIC) design service factory Creative, now successfully developing the HBM key peripheral components required for AI servers, jointly inking a "large order" for base dies required for next-gen HBM4 memory.

UDN's report says that TSMC and Creative have "never commented on order developments," but the legal source of the site said that demand for AI is strong, and not just HPC-related chips, but HBM demand is "growing rapidly" and is "becoming a new business opportunity in the market" which has three major memory chip companies including SK hynix, Samsung, and Micron, investing as much as they can into their respective HBM supply chains.