Indian Express calls it "the ultimate smartphone killer". (Coming soon, its laser-on-your-palm feature will display stock prices, sports scores, and flight statuses.)

Humane's Ai Pin can even translate what you say, repeating it out loud in another language (with 50 different languages supported). And it can read you summaries of what's on your favorite web sites, so "You can just surf the web with your voice," according to a new video released this week.

The video also shows it answering specific questions like "What's that song by 21 Savage with the violin intro?" (And later, while the song is playing, answering more questions like "This was sampled from another song. What song was that?") But then co-founder Imran Chaudhri — an iPhone designer and one of several former Apple employees at Humane — demonstrated a "Vision" feature that's coming soon. Holding a Sony Walkman he asks the Pin to "Look at this and tell me when it first came out" — and the Pin obliges. ("The Sony Walkman WM-F73 was released in 1986...") In another demo it correctly supplied the designer of an Air Jordan basketball shoe.

They're also working on integrating this into a Nutrition Tracking application. (A demonstrator held a doughnut and asked the Pin to identify how much sugar was in it.) If you tell the Pin that you've eaten the doughnut, it can then calculate your intake of carbs, protein, and fats.

And in the video the Pin responded within seconds to the command "Make a spreadsheet about top consumer tech reviewers on YouTube [with] real names, subscriber counts, and URLs." It performed the research and created the spreadsheet, which appears on the demonstrator's laptop, apparently logged in to Humane's cloud-based user platform.

In the video Humane's co-founder stresses that its Ai Pin does all this without downloading applications, "which allows me to stay present in the moment and flow." But while it can also make phone calls and sends text messages, Imran Chaudhri adds that "Ai Pin is a completely new form factor for compute. It's never been about replacing. It's always been about creating new ways to interact with what you need. So instead of having to sit down to use a computer, or reaching in to your pocket and pulling out your phone and navigating apps, Ai Pin allows you to simply act on something the moment you think about it — letting AI do all the work for you."

Or, as they say later "This is about technology adapting and reacting to you. Not you having to adapt to it."

There's also talk about their "AI OS" — named Cosmos — with the Pin described as "our first entry point" into that operating system, with other devices planned to support it in the future. (Mashable's reporter notes that Humane's Ai Pin is backed by OpenAI CEO Sam Altman, and writes "I was impressed with how well it worked.") The video even ends with an update for SDK developers. In the second half of 2024, "you're going to be able to connect your services to the Ai Pin using REST APIs and OAuth." Phase two will let developers run their code directly on Humane's cloud platform — while Phase three will see developers codes on Ai Pin devices, "to get access to the mic, the camera, the sensors, and the laser. We are so excited to see what you're gonna build."

Humane says its Ai Pin will start shipping at the end of March, with priority orders arriving starting on April 11th.

An AI tool named Mia, tested by the NHS, successfully detected signs of breast cancer in 11 women which had been missed by human doctors. The BBC reports: The tool, called Mia, was piloted alongside NHS clinicians and analyzed the mammograms of over 10,000 women. Most of them were cancer-free, but it successfully flagged all of those with symptoms, as well as an extra 11 the doctors did not identify. At their earliest stages, cancers can be extremely small and hard to spot. The BBC saw Mia in action at NHS Grampian, where we were shown tumors that were practically invisible to the human eye. But, depending on their type, they can grow and spread rapidly.

Barbara was one of the 11 patients whose cancer was flagged by Mia but had not been spotted on her scan when it was studied by the hospital radiologists. Because her 6mm tumor was caught so early she had an operation but only needed five days of radiotherapy. Breast cancer patients with tumors which are smaller than 15mm when discovered have a 90% survival rate over the following five years. Barbara said she was pleased the treatment was much less invasive than that of her sister and mother, who had previously also battled the disease. Without the AI tool's assistance, Barbara's cancer would potentially not have been spotted until her next routine mammogram three years later. She had not experienced any noticeable symptoms. "These results are encouraging and help to highlight the exciting potential AI presents for diagnostics. There is no question that real-life clinical radiologists are essential and irreplaceable, but a clinical radiologist using insights from validated AI tools will increasingly be a formidable force in patient care." said Dr Katharine Halliday, President of the Royal College of Radiologists.

Search Engine Land: Google is now testing AI overviews in the main Google Search results, even if you have not opted into the Google Search Generative Experience labs feature. Google said this is an experience on a "subset of queries, on a small percentage of search traffic in the U.S.," a Google spokesperson told Search Engine Land.

Apple has held preliminary talks with Baidu about using the Chinese company's generative AI technology in its devices in China, the latest example of the iPhone maker's efforts to widen its AI capabilities. From a report: The U.S. tech giant has been exploring using external partners to help accelerate its AI ambitions. It has held discussions with companies including Google and OpenAI about using their technology to power its mobile features. In China, Apple has been looking for a local generative AI model provider, mainly because China requires such models to be vetted by its cyberspace regulator before being launched to the public, people familiar with the matter said.

Longtime Slashdot reader jgulla shares an announcement from Redis: Beginning today, all future versions of Redis will be released with source-available licenses. Starting with Redis 7.4, Redis will be dual-licensed under the Redis Source Available License (RSALv2) and Server Side Public License (SSPLv1). Consequently, Redis will no longer be distributed under the three-clause Berkeley Software Distribution (BSD). The new source-available licenses allow us to sustainably provide permissive use of our source code.

We're leading Redis into its next phase of development as a real-time data platform with a unified set of clients, tools, and core Redis product offerings. The Redis source code will continue to be freely available to developers, customers, and partners through Redis Community Edition. Future Redis source-available releases will unify core Redis with Redis Stack, including search, JSON, vector, probabilistic, and time-series data models in one free, easy-to-use package as downloadable software. This will allow anyone to easily use Redis across a variety of contexts, including as a high-performance key/value and document store, a powerful query engine, and a low-latency vector database powering generative AI applications. [...]

Under the new license, cloud service providers hosting Redis offerings will no longer be permitted to use the source code of Redis free of charge. For example, cloud service providers will be able to deliver Redis 7.4 only after agreeing to licensing terms with Redis, the maintainers of the Redis code. These agreements will underpin support for existing integrated solutions and provide full access to forthcoming Redis innovations. In practice, nothing changes for the Redis developer community who will continue to enjoy permissive licensing under the dual license. At the same time, all the Redis client libraries under the responsibility of Redis will remain open source licensed. Redis will continue to support its vast partner ecosystem -- including managed service providers and system integrators -- with exclusive access to all future releases, updates, and features developed and delivered by Redis through its Partner Program. There is no change for existing Redis Enterprise customers.

An anonymous reader quotes a report from Reuters: The United Nations General Assembly on Thursday unanimously adopted the first global resolution on artificial intelligence to encourage protecting personal data, monitoring AI for risks, and safeguarding human rights, U.S. officials said. The nonbinding resolution, proposed by the United States and co-sponsored by China and 121 other nations, took three months to negotiate and also advocates strengthening privacy policies, the officials said, briefing reporters before the resolution's passage. "We're sailing in choppy waters with the fast-changing technology, which means that its more important than ever to steer by the light of our values," said one of the senior administration officials, describing the resolution as the "first-ever truly global consensus document on AI."

"The improper or malicious design, development, deployment and use of artificial intelligence systems ... pose risks that could ... undercut the protection, promotion and enjoyment of human rights and fundamental freedoms," the measure says. Asked whether negotiators faced resistance from Russia or China -- U.N. member states that also voted in favor of the document -- the officials conceded there were "lots of heated conversations. ... But we actively engaged with China, Russia, Cuba, other countries that often don't see eye to eye with us on issues." "We believe the resolution strikes the appropriate balance between furthering development, while continuing to protect human rights," said one of the officials, who spoke on condition of anonymity.

Microsoft has announced two new Surface devices, the Surface Pro 10 for Business and Surface Laptop 6 for Business, both featuring Intel's latest Core Ultra processors, a dedicated Neural Processing Unit (NPU), and a new Copilot key for AI-powered features in Windows 11.

The devices, which will start shipping to commercial customers on April 9th, have been designed exclusively for businesses and will not be sold directly to consumers. The Surface Pro 10 for Business, starting at $1,199, offers a choice between Core Ultra 5 135U and Core Ultra 7 165U options, with up to 64GB of RAM and a 256GB Gen4 SSD. It also features an improved 13-inch display with an antireflective coating and a 1440p front-facing camera with a 114-degree field of view.

The Surface Laptop 6 for Business, also starting at $1,199, is powered by Intel's Core Ultra H-series chips and is available with up to 64GB of RAM and a 1TB Gen4 SSD. The 15-inch model includes two USB-C Thunderbolt 4 ports, while the 13.5-inch model features a single USB-C Thunderbolt 4 port. Both devices have an optional smart card reader and are Microsoft's most easily serviceable Surface devices to date.

Further reading: Microsoft's official blog.

After securing billions in federal grants and loans, Reuters reports that the company is "planning a $100-billion spending spree across four U.S. states" to build and expand its chip manufacturing factories. From the report: The centerpiece of Intel's five-year spending plan is turning empty fields near Columbus, Ohio, into what CEO Pat Gelsinger described to reporters on Tuesday as "the largest AI chip manufacturing site in the world," starting as soon as 2027. Intel's plan will also involve revamping sites in New Mexico and Oregon and expanding operations in Arizona, where longtime rival Taiwan Semiconductor Manufacturing Co is also building a massive factory that it hopes will receive funding from President Joe Biden's push to bring advanced semiconductor manufacturing back to the United States. [...]

Gelsinger said about 30% of the $100-billion plan will be spent on construction costs such as labor, piping and concrete. The remaining will go towards buying chipmaking tools from firms such as ASML, Tokyo Electron, Applied Materials and KLA, among others. Those tools will help bring the Ohio site online by 2027 or 2028, though Gelsinger warned the timeline could slip if the chip market takes a dive. Beyond grants and loans, Intel plans to make most of the purchases from its existing cash flows.

"It will still take three to five years for Intel to become a serious player in the foundry market" for cutting-edge chips, said Kinngai Chan, an analyst at Summit Insights. However, he warned more investment would be needed before Intel could overtake TSMC, adding that the Taiwanese firm could remain the leader for "some time to come." Gelsinger has previously said a second round of U.S. funding for chip factories would likely be needed to re-establish the U.S. as a leader in semiconductor manufacturing, which he reiterated on Tuesday. "It took us three-plus decades to lose this industry. It's not going to come back in three to five years of CHIPS Act" funding, said Gelsinger, who referred to the low-interest-rate funding as "smart capital."

An anonymous reader quotes a report from Ars Technica: Google closed its Fitbit acquisition in 2021. Since then, the tech behemoth has pushed numerous changes to the wearable brand, including upcoming updates announced this week. While Google reshapes its fitness tracker business, though, some long-time users are regretting their Fitbit purchases and questioning if Google's practices will force them to purchase their next fitness tracker elsewhere.

As is becoming common practice with consumer tech announcements of late, Google's latest announcements about Fitbit seemed to be trying to convince users of the wonders of generative AI and how that will change their gadgets for the better. In a blog post yesterday, Dr. Karen DeSalvo, Google's chief health officer, announced that Fitbit Premium subscribers would be able to test experimental AI features later this year (Google hasn't specified when). "You will be able to ask questions in a natural way and create charts just for you to help you understand your own data better. For example, you could dig deeper into how many active zone minutes... you get and the correlation with how restorative your sleep is," she wrote. DeSalvo's post included an example of a user asking a chatbot if there was a connection between their sleep and activity and said that the experimental AI features will only be available to "a limited number of Android users who are enrolled in the Fitbit Labs program in the Fitbit mobile app."

Fitbit is also working with the Google Research team and "health and wellness experts, doctors, and certified coaches" to develop a large language model (LLM) for upcoming Fitbit mobile app features that pull data from Fitbit and Pixel devices, DeSalvo said. In a blog post yesterday, Yossi Matias, VP of engineering and research at Google, said Google wants to use the LLM to add personalized coaching features, such as the ability to look for sleep irregularities and suggest actions "on how you might change the intensity of your workout." Google's Fitbit is building the LLM on Gemini models that are tweaked on de-identified data from unspecified "research case studies," Matias said, adding: "For example, we're testing performance using sleep medicine certification exam-like practice tests." Other recent changes to Fitbit include a name tweak from Fitbit by Google, to Google Fitbit, as spotted by 9to5Google this week. Charge 5 users are especially concerned after users noticed their devices suddenly stopped holding a charge after a December firmware update was pushed. The problem has persisted with Google offering no solution other than offer discounts or, if the device was within its warranty period, a replacement.

"This is called planned obsolescence. I'll be upgrading to a watch style tracker from a different company. I wish Fitbit hadn't sold out to Google," a forum user going by Sean77024 wrote on Fitbit's support forum yesterday. "Others, like 2MeFamilyFlyer, have also accused Fitbit of planning Charge 5 obsolescence," notes Ars. "2MeFamilyFlyer said they're seeking a Fitbit alternative."

NVIDIA has been working on adding generative AI to non-playable characters (NPCs) for a while now. The company is hoping a newly-announced partnership with Ubisoft will accelerate development of this technology and, ultimately, bring these AI-driven NPCs to modern games. From a report: Ubisoft helped build new "NEO NPCs" by using NVIDIA's Avatar Cloud Engine (ACE) technology, with an assist from dynamic NPC experts Inworld AI. The end result? Characters that don't repeat the same phrase over and over, while ignoring the surrounding violent mayhem. These NEO NPCs are said to interact in real time with players, the environment and other in-game characters. NVIDIA says this opens up "new possibilities for emergent storytelling." To that end, Ubisoft's narrative team built complete backgrounds, knowledge bases and conversational styles for two NPCs as a proof of concept.

An anonymous reader shares a report: When OpenAI CEO Sam Altman announced GPTs, custom chatbots powered by OpenAI's generative AI models, onstage at the company's first-ever developer conference in November, he described them as a way to "accomplish all sorts of tasks" -- from programming to learning about esoteric scientific subjects to getting workout pointers. "Because [GPTs] combine instructions, expanded knowledge and actions, they can be more helpful to you," Altman said. "You can build a GPT ... for almost anything." He wasn't kidding about the anything part.

TechCrunch found that the GPT Store, OpenAI's official marketplace for GPTs, is flooded with bizarre, potentially copyright-infringing GPTs that imply a light touch where it concerns OpenAI's moderation efforts. A cursory search pulls up GPTs that purport to generate art in the style of Disney and Marvel properties, serve as little more than funnels to third-party paid services, advertise themselves as being able to bypass AI content detection tools such as Turnitin and Copyleaks.

An anonymous reader shares a report: The generative AI company helmed by Sam Altman is on track to put out GPT-5 sometime mid-year, likely during summer, according to two people familiar with the company. Some enterprise customers have recently received demos of the latest model and its related enhancements to the ChatGPT tool, another person familiar with the process said. These people, whose identities Business Insider has confirmed, asked to remain anonymous so they could speak freely. "It's really good, like materially better," said one CEO who recently saw a version of GPT-5.

OpenAI demonstrated the new model with use cases and data unique to his company, the CEO said. He said the company also alluded to other as-yet-unreleased capabilities of the model, including the ability to call AI agents being developed by OpenAI to perform tasks autonomously. The company does not yet have a set release date for the new model, meaning current internal expectations for its release could change. OpenAI is still training GPT-5, one of the people familiar said. After training is complete, it will be safety tested internally and further "red teamed," a process where employees and typically a selection of outsiders challenge the tool in various ways to find issues before it's made available to the public.

Haje Jan Kamps writes via TechCrunch: Artificial General Intelligence (AGI) -- often referred to as "strong AI," "full AI," "human-level AI" or "general intelligent action" -- represents a significant future leap in the field of artificial intelligence. Unlike narrow AI, which is tailored for specific tasks (such as detecting product flaws, summarize the news, or build you a website), AGI will be able to perform a broad spectrum of cognitive tasks at or above human levels. Addressing the press this week at Nvidia's annual GTC developer conference, CEO Jensen Huang appeared to be getting really bored of discussing the subject -- not least because he finds himself misquoted a lot, he says. The frequency of the question makes sense: The concept raises existential questions about humanity's role in and control of a future where machines can outthink, outlearn and outperform humans in virtually every domain. The core of this concern lies in the unpredictability of AGI's decision-making processes and objectives, which might not align with human values or priorities (a concept explored in depth in science fiction since at least the 1940s). There's concern that once AGI reaches a certain level of autonomy and capability, it might become impossible to contain or control, leading to scenarios where its actions cannot be predicted or reversed.

When sensationalist press asks for a timeframe, it is often baiting AI professionals into putting a timeline on the end of humanity -- or at least the current status quo. Needless to say, AI CEOs aren't always eager to tackle the subject. Predicting when we will see a passable AGI depends on how you define AGI, Huang argues, and draws a couple of parallels: Even with the complications of time-zones, you know when new year happens and 2025 rolls around. If you're driving to the San Jose Convention Center (where this year's GTC conference is being held), you generally know you've arrived when you can see the enormous GTC banners. The crucial point is that we can agree on how to measure that you've arrived, whether temporally or geospatially, where you were hoping to go. "If we specified AGI to be something very specific, a set of tests where a software program can do very well -- or maybe 8% better than most people -- I believe we will get there within 5 years," Huang explains. He suggests that the tests could be a legal bar exam, logic tests, economic tests or perhaps the ability to pass a pre-med exam. Unless the questioner is able to be very specific about what AGI means in the context of the question, he's not willing to make a prediction. Fair enough.

An anonymous reader quotes a report from Wired: YouTube has updated its rulebook for the era of deepfakes. Starting today, anyone uploading video to the platform must disclose certain uses of synthetic media, including generative AI, so viewers know what they're seeing isn't real. YouTube says it applies to "realistic" altered media such as "making it appear as if a real building caught fire" or swapping"the face of one individual with another's." The new policy shows YouTube taking steps that could help curb the spread of AI-generated misinformation as the US presidential election approaches. It is also striking for what it permits: AI-generated animations aimed at kids are not subject to the new synthetic content disclosure rules.

YouTube's new policies exclude animated content altogether from the disclosure requirement. This means that the emerging scene of get-rich-quick, AI-generated content hustlers can keep churning out videos aimed at children without having to disclose their methods. Parents concerned about the quality of hastily made nursery-rhyme videos will be left to identify AI-generated cartoons by themselves. YouTube's new policy also says creators don't need to flag use of AI for "minor" edits that are "primarily aesthetic" such as beauty filters or cleaning up video and audio. Use of AI to "generate or improve" a script or captions is also permitted without disclosure. [...]

The exemption for animation in YouTube's new policy could mean that parents cannot easily filter such videos out of search results or keep YouTube's recommendation algorithm from autoplaying AI-generated cartoons after setting up their child to watch popular and thoroughly vetted channels like PBS Kids or Ms. Rachel. Some problematic AI-generated content aimed at kids does require flagging under the new rules. In 2023, the BBC investigated a wave of videos targeting older children that used AI tools to push pseudoscience and conspiracy theories, including climate change denialism. These videos imitated conventional live-action educational videos -- showing, for example, the real pyramids of Giza -- so unsuspecting viewers might mistake them for factually accurate educational content. (The pyramid videos then went on the suggest that the structures can generate electricity.) This new policy would crack down on that type of video. "We require kids content creators to disclose content that is meaningfully altered or synthetically generated when it seems realistic," says YouTube spokesperson Elena Hernandez. "We don't require disclosure of content that is clearly unrealistic and isn't misleading the viewer into thinking it's real."

According to the New York Times, Saudi Arabia's government plans to create a fund of about $40 billion to invest in artificial intelligence. Reuters reports: Representatives of Saudi Arabia's Public Investment Fund (PIF) have discussed a potential partnership with U.S. venture capital firm Andreessen Horowitz and other financiers in recent weeks, the newspaper reported. Andreessen Horowitz and PIF governor Yasir Al-Rumayyan have discussed the possibility of the U.S. firm setting up an office in Riyadh, according to the report. PIF officials also discussed what role Andreessen Horowitz could play and how such a fund would work, the newspaper said, adding the plans could still change. Other venture capitalists may participate in kingdom's artificial intelligence fund, which is expected to commence in the second half of 2024, the newspaper said.

Saudi representatives have indicated to potential partners that the country is interested in supporting a variety of tech start-ups associated with artificial intelligence, including chip makers and large-scale data centers, the report added. Last month, PIF's Al-Rumayyan pitched the kingdom as a prospective hub for artificial intelligence activity outside U.S., citing its energy resources and funding capacity. Al-Rumayyan had said the kingdom had the "political will" to make artificial intelligence projects happen and ample funds it could deploy to nurture the technology's development.

Academics in the artificial intelligence field have started using generative AI services to help them review the machine learning work of their peers. In a new paper on arXiv, researchers analyzed the peer reviews of papers submitted to leading AI conferences, including ICLR 2024, NeurIPS 2023, CoRL 2023 and EMNLP 2023. The Register reports on the findings: The authors took two sets of data, or corpora -- one written by humans and the other one written by machines. And they used these two bodies of text to evaluate the evaluations -- the peer reviews of conference AI papers -- for the frequency of specific adjectives. "[A]ll of our calculations depend only on the adjectives contained in each document," they explained. "We found this vocabulary choice to exhibit greater stability than using other parts of speech such as adverbs, verbs, nouns, or all possible tokens."

It turns out LLMs tend to employ adjectives like "commendable," "innovative," and "comprehensive" more frequently than human authors. And such statistical differences in word usage have allowed the boffins to identify reviews of papers where LLM assistance is deemed likely. "Our results suggest that between 6.5 percent and 16.9 percent of text submitted as peer reviews to these conferences could have been substantially modified by LLMs, i.e. beyond spell-checking or minor writing updates," the authors argued, noting that reviews of work in the scientific journal Nature do not exhibit signs of mechanized assistance. Several factors appear to be correlated with greater LLM usage. One is an approaching deadline: The authors found a small but consistent increase in apparent LLM usage for reviews submitted three days or less before the deadline.

The researchers emphasized that their intention was not to pass judgment on the use of AI writing assistance, nor to claim that any of the papers they evaluated were written completely by an AI model. But they argued the scientific community needs to be more transparent about the use of LLMs. And they contended that such practices potentially deprive those whose work is being reviewed of diverse feedback from experts. What's more, AI feedback risks a homogenization effect that skews toward AI model biases and away from meaningful insight.

Soccer teams are always looking to get an edge over their rivals. Whether it's studying players' susceptibility to injury, or opponents' tactics -- top clubs look at reams of data to give them the best shot of winning. They might want to add a new AI assistant developed by Google DeepMind to their arsenal. From a report: It can suggest tactics for soccer set-pieces that are even better than those created by professional club coaches. The system, called TacticAI, works by analyzing a dataset of 7,176 corner kicks taken by players for Liverpool FC, one of the biggest soccer clubs in the world. Corner kicks are awarded to an attacking team when the ball passes over the goal line after touching a player on the defending team. In a sport as free-flowing and unpredictable as soccer, corners -- like free kicks and penalties -- are rare instances in the game when teams can try out pre-planned plays.

TacticAI uses predictive and generative AI models to convert each corner kick scenario -- such as a receiver successfully scoring a goal, or a rival defender intercepting the ball and returning it to their team -- into a graph, and the data from each player into a node on the graph, before modeling the interactions between each node. The work was published in Nature Communications today. Using this data, the model provides recommendations about where to position players during a corner to give them, for example, the best shot at scoring a goal, or the best combination of players to get up front. It can also try to predict the outcomes of a corner, including whether a shot will take place, or which player is most likely to touch the ball first.

Microsoft has named Mustafa Suleyman head of its consumer artificial intelligence business, hiring most of the staff from his Inflection AI startup as the software giant seeks to fend off Alphabet's Google in the fiercely contested market for AI products. From a report: Suleyman, who co-founded Google's DeepMind, will report to Chief Executive Officer Satya Nadella and oversee a range of projects, such as integrating an AI Copilot into Windows and adding conversational elements to the Bing search engine. His hiring will put Microsoft's consumer AI work under one leader for the first time.

Inflection, a rival of Microsoft's key AI partner OpenAI, is exiting its Pi consumer chatbot effort and shifting to selling AI software to businesses. Karen Simonyan, Inflection's co-founder, will join Microsoft as chief scientist for the new consumer AI group. In the past year, Nadella has been revamping his company's major products around artificial intelligence technology from OpenAI. Under the Copilot brand, Microsoft has blended an AI assistant into products including Windows, consumer and enterprise Office software, Bing and security tools. With Google and others trying to catch up, Nadella's multibillion-dollar investment in OpenAI has given Microsoft a first-mover advantage. And yet, 13 months after unveiling an AI-enhanced Bing search, the company has made few gains in that market, which remains dominated by Google.

Sean Hollister reports via The Verge: Nvidia's must-have H100 AI chip made it a multitrillion-dollar company, one that may be worth more than Alphabet and Amazon, and competitors have been fighting to catch up. But perhaps Nvidia is about to extend its lead -- with the new Blackwell B200 GPU and GB200 "superchip." Nvidia says the new B200 GPU offers up to 20 petaflops of FP4 horsepower from its 208 billion transistors and that a GB200 that combines two of those GPUs with a single Grace CPU can offer 30 times the performance for LLM inference workloads while also potentially being substantially more efficient. It "reduces cost and energy consumption by up to 25x" over an H100, says Nvidia.

Training a 1.8 trillion parameter model would have previously taken 8,000 Hopper GPUs and 15 megawatts of power, Nvidia claims. Today, Nvidia's CEO says 2,000 Blackwell GPUs can do it while consuming just four megawatts. On a GPT-3 LLM benchmark with 175 billion parameters, Nvidia says the GB200 has a somewhat more modest seven times the performance of an H100, and Nvidia says it offers 4x the training speed. Nvidia told journalists one of the key improvements is a second-gen transformer engine that doubles the compute, bandwidth, and model size by using four bits for each neuron instead of eight (thus, the 20 petaflops of FP4 I mentioned earlier). A second key difference only comes when you link up huge numbers of these GPUs: a next-gen NVLink switch that lets 576 GPUs talk to each other, with 1.8 terabytes per second of bidirectional bandwidth. That required Nvidia to build an entire new network switch chip, one with 50 billion transistors and some of its own onboard compute: 3.6 teraflops of FP8, says Nvidia. Further reading: Nvidia in Talks To Acquire AI Infrastructure Platform Run:ai

Cisco on Monday completed its $28 billion acquisition of Splunk, a powerhouse in data analysis, security and observability tools. The deal was first announced in September 2023. SecurityWeek reports: Cisco plans to leverage Splunk's AI, security and observability capabilities complement Cisco's solution portfolio. Cisco says the transaction is expected to be cash flow positive and non-GAAP gross margin accretive in Cisco's fiscal year 2025, and non-GAAP EPS accretive in fiscal year 2026. "We are thrilled to officially welcome Splunk to Cisco," Chuck Robbins, Chair and CEO of Cisco, said in a statement. "As one of the world's largest software companies, we will revolutionize the way our customers leverage data to connect and protect every aspect of their organization as we help power and protect the AI revolution."