Author name: Kris Guyer

ars-drives-the-second-generation-rivian-r1t-and-r1s-electric-trucks

Ars drives the second-generation Rivian R1T and R1S electric trucks

car review, Cars, electric truck, Rivian, Rivian R1S, Rivian R1T / /

no more car sick —

The EV startup has reengineered the R1 to make it better to drive, easier to build.

–

A Rivian R1T and R1S parked together in a forest

Enlarge / The R1S and R1T don’t look much different from the electric trucks we drove in 2022, but under the skin, there have been a lot of changes.

Rivian

In rainy Seattle this week, Rivian unveiled what it’s calling the “Second Generation” of its R1 line with a suite of mostly under-the-hood software and hardware updates that increase range, power, and efficiency while simultaneously lowering the cost of production for the company. While it’s common for automotive manufacturers to do some light refreshes after about four model years, Rivian has almost completely retooled the underpinnings of its popular R1S SUV and R1T pickup just two years after the vehicles made their debut.

“Overdelivering on the product is one of our core values,” Wassym Bensaid, the chief software officer at Rivian, told a select group of journalists at the event on Monday night, “and customer feedback has been one of the key inspirations for us.”

For these updates, Rivian changed more than half the hardware components in the R1 platform, retooled its drive units to offer new tri- and quad-motor options (with more horsepower), updated the suspension tuning, deleted 1.6 miles (2.6 km) of wiring, reduced the number of ECUs, increased the number of cameras and sensors around the vehicle, changed the battery packs, and added some visual options that better aligned with customizations that owners were making to their vehicles, among other things. Rivian is also leaning harder into AI and ML tools with the aim of bringing limited hands-free driver-assistance systems to their owners toward the end of the year.

  • Usually, an automaker waits four years before it refreshes a product, but Rivian decided to move early.

    Rivian

  • The R1 interior can feel quite serene.

    Rivian

  • Perhaps you’d prefer something more colorful?

    Rivian

  • An exploded view of a drive unit with a pair of motors.

    Rivian

  • There are two capacities of lithium-ion battery, and an optional lithium iron phosphate pack with 275 miles of range is on the way.

  • Rivian’s R1 still looks friendly amid a sea of scary-looking SUVs and trucks.

    Rivian

While many of these changes have simplified manufacturing for Rivian, which as of Q1 of this year lost a whopping $38,000 on every vehicle it sold, the company has continued to close the gap with the likes of BMW and Mercedes in terms of ride, handling, comfort, and efficiency.

On the road in the new R1

We drove a new second-gen dual-motor 665 hp (496 kW), 829 lb-ft (1,124 Nm) R1S Performance, which gets up to 410 miles (660 km) of range with the new Max Pack battery, out to DirtFish Rally School in Snoqualmie in typically rainy Seattle weather. On the road, the new platform, with its revised suspension and shocks, felt much more comfortable than it did in our first experience with an R1S in New York in 2022.

The vehicle offers modes that allow you to tackle pretty much any kind of driving that life can throw at you, including Sport, All Purpose (there’s no longer a “Conserve” mode), Snow, All-Terrain, and Soft Sand, alongside customizable suspension, ride feel and height, and regen settings. The R1S feels far more comfortable from all seating positions, including the back and third-row seats. There’s less floaty, car-sick-inducing modulation over bumps in All-Purpose, and Sport tightens things down nicely when you want to have a bit more road feel.

One of the big improvements on the road comes from the new “Autonomy Compute Module” and its suite of high-resolution 4K HDR cameras, radars, and sensors that have been upgraded on the R1 platform. The new R1 gets 11 cameras (one more than the first gen), with eight times greater resolution, five radar modules, and a new proprietary AI and ML integrated system that learns from anonymized driver data and information taken from the world around the vehicles to “see” 360-degrees around the vehicle, even in inclement weather.

While the R1S has had cruise control since its launch, the new “Autonomy” platform allows for smart lane-changing—something Rivian calls “Lane Change on Command” when using the new “Enhanced Highway Assist” (a partially automated driver assist), and centers the vehicle in marked lanes. We tried both features on the highways around Seattle, and the system handled very rainy and wet weather without hesitation, but it did ping-pong between the lane markers, and when that smart lane change system bailed out at the last minute, the move was abrupt and not confidence-inspiring, since there was no apparent reason for the system to fail. These features are not nearly as good as the latest from BMW and Mercedes, both of which continue to offer some of the most usable driver-assist systems on the market.

With the new R1 software stack, Rivian is also promising some limited hands-free highway driver-assistance features to come at the end of the year. While we didn’t get to try the feature in the short drive to DirtFish, Rivian says eye-tracking cameras in the rearview mirror will ensure that drivers have ample warning to take over when the system is engaged and needs human input.

Ars drives the second-generation Rivian R1T and R1S electric trucks Read More »

can-a-technology-called-rag-keep-ai-models-from-making-stuff-up?

Can a technology called RAG keep AI models from making stuff up?

AI, artificial inteligence, confabulation, Features, generative ai, LLMs, RAG, retrieval augmented generation / /
Can a technology called RAG keep AI models from making stuff up?

Aurich Lawson | Getty Images

We’ve been living through the generative AI boom for nearly a year and a half now, following the late 2022 release of OpenAI’s ChatGPT. But despite transformative effects on companies’ share prices, generative AI tools powered by large language models (LLMs) still have major drawbacks that have kept them from being as useful as many would like them to be. Retrieval augmented generation, or RAG, aims to fix some of those drawbacks.

Perhaps the most prominent drawback of LLMs is their tendency toward confabulation (also called “hallucination”), which is a statistical gap-filling phenomenon AI language models produce when they are tasked with reproducing knowledge that wasn’t present in the training data. They generate plausible-sounding text that can veer toward accuracy when the training data is solid but otherwise may just be completely made up.

Relying on confabulating AI models gets people and companies in trouble, as we’ve covered in the past. In 2023, we saw two instances of lawyers citing legal cases, confabulated by AI, that didn’t exist. We’ve covered claims against OpenAI in which ChatGPT confabulated and accused innocent people of doing terrible things. In February, we wrote about Air Canada’s customer service chatbot inventing a refund policy, and in March, a New York City chatbot was caught confabulating city regulations.

So if generative AI aims to be the technology that propels humanity into the future, someone needs to iron out the confabulation kinks along the way. That’s where RAG comes in. Its proponents hope the technique will help turn generative AI technology into reliable assistants that can supercharge productivity without requiring a human to double-check or second-guess the answers.

“RAG is a way of improving LLM performance, in essence by blending the LLM process with a web search or other document look-up process” to help LLMs stick to the facts, according to Noah Giansiracusa, associate professor of mathematics at Bentley University.

Let’s take a closer look at how it works and what its limitations are.

A framework for enhancing AI accuracy

Although RAG is now seen as a technique to help fix issues with generative AI, it actually predates ChatGPT. Researchers coined the term in a 2020 academic paper by researchers at Facebook AI Research (FAIR, now Meta AI Research), University College London, and New York University.

As we’ve mentioned, LLMs struggle with facts. Google’s entry into the generative AI race, Bard, made an embarrassing error on its first public demonstration back in February 2023 about the James Webb Space Telescope. The error wiped around $100 billion off the value of parent company Alphabet. LLMs produce the most statistically likely response based on their training data and don’t understand anything they output, meaning they can present false information that seems accurate if you don’t have expert knowledge on a subject.

LLMs also lack up-to-date knowledge and the ability to identify gaps in their knowledge. “When a human tries to answer a question, they can rely on their memory and come up with a response on the fly, or they could do something like Google it or peruse Wikipedia and then try to piece an answer together from what they find there—still filtering that info through their internal knowledge of the matter,” said Giansiracusa.

But LLMs aren’t humans, of course. Their training data can age quickly, particularly in more time-sensitive queries. In addition, the LLM often can’t distinguish specific sources of its knowledge, as all its training data is blended together into a kind of soup.

In theory, RAG should make keeping AI models up to date far cheaper and easier. “The beauty of RAG is that when new information becomes available, rather than having to retrain the model, all that’s needed is to augment the model’s external knowledge base with the updated information,” said Peterson. “This reduces LLM development time and cost while enhancing the model’s scalability.”

Can a technology called RAG keep AI models from making stuff up? Read More »

elon-musk’s-x-defeats-australia’s-global-takedown-order-of-stabbing-video

Elon Musk’s X defeats Australia’s global takedown order of stabbing video

Australian, censorship, elon musk, First Amendment, free speech, julie inman-grant, online safety act, Policy, Twitter, X, x corp / /
Elon Musk’s X defeats Australia’s global takedown order of stabbing video

Australia’s safety regulator has ended a legal battle with X (formerly Twitter) after threatening approximately $500,000 daily fines for failing to remove 65 instances of a religiously motivated stabbing video from X globally.

Enforcing Australia’s Online Safety Act, eSafety commissioner Julie Inman-Grant had argued it would be dangerous for the videos to keep spreading on X, potentially inciting other acts of terror in Australia.

But X owner Elon Musk refused to comply with the global takedown order, arguing that it would be “unlawful and dangerous” to allow one country to control the global Internet. And Musk was not alone in this fight. The legal director of a nonprofit digital rights group called the Electronic Frontier Foundation (EFF), Corynne McSherry, backed up Musk, urging the court to agree that “no single country should be able to restrict speech across the entire Internet.”

“We welcome the news that the eSafety Commissioner is no longer pursuing legal action against X seeking the global removal of content that does not violate X’s rules,” X’s Global Government Affairs account posted late Tuesday night. “This case has raised important questions on how legal powers can be used to threaten global censorship of speech, and we are heartened to see that freedom of speech has prevailed.”

Inman-Grant was formerly Twitter’s director of public policy in Australia and used that experience to land what she told The Courier-Mail was her “dream role” as Australia’s eSafety commissioner in 2017. Since issuing the order to remove the video globally on X, Inman-Grant had traded barbs with Musk (along with other Australian lawmakers), responding to Musk labeling her a “censorship commissar” by calling him an “arrogant billionaire” for fighting the order.

On X, Musk arguably got the last word, posting, “Freedom of speech is worth fighting for.”

Safety regulator still defends takedown order

In a statement, Inman-Grant said early Wednesday that her decision to discontinue proceedings against X was part of an effort to “consolidate actions,” including “litigation across multiple cases.” She ultimately determined that dropping the case against X would be the “option likely to achieve the most positive outcome for the online safety of all Australians, especially children.”

“Our sole goal and focus in issuing our removal notice was to prevent this extremely violent footage from going viral, potentially inciting further violence and inflicting more harm on the Australian community,” Inman-Grant said, still defending the order despite dropping it.

In court, X’s lawyer Marcus Hoyne had pushed back on such logic, arguing that the eSafety regulator’s mission was “pointless” because “footage of the attack had now spread far beyond the few dozen URLs originally identified,” the Australian Broadcasting Corporation reported.

“I stand by my investigators and the decisions eSafety made,” Inman-Grant said.

Other Australian lawmakers agree the order was not out of line. According to AP News, Australian Minister for Communications Michelle Rowland shared a similar statement in parliament today, backing up the safety regulator while scolding X users who allegedly took up Musk’s fight by threatening Inman-Grant and her family. The safety regulator has said that Musk’s X posts incited a “pile-on” from his followers who allegedly sent death threats and exposed her children’s personal information, the BBC reported.

“The government backs our regulators and we back the eSafety Commissioner, particularly in light of the reprehensible threats to her physical safety and the threats to her family in the course of doing her job,” Rowland said.

Elon Musk’s X defeats Australia’s global takedown order of stabbing video Read More »

the-challenge-of-securing-user-identities

The Challenge of Securing User Identities

Challenge / /

Several businesses I’ve worked with recently have had the misfortune of being victims of cybersecurity incidents. While these incidents come in many forms, there is a common thread: they all started with a compromise of user identity.

Why Identities are Targeted

Identity security—whether it involves usernames and passwords, machine names, encryption keys, or certificates—presents a real challenge. These credentials are needed for access control, ensuring only authorized users have access to systems, infrastructure, and data. Cybercriminals also know this, which is why they are constantly trying to compromise credentials. It’s why incidents such as phishing attacks remain an ongoing problem; gaining access to the right credentials is the foothold an attacker needs.

Attempts to compromise identity do leave a trail: a phishing email, an attempted logon from an incorrect location, or more sophisticated signs such as the creation of a new multifactor authentication (MFA) token. Unfortunately, these things can happen many days apart, are often recorded across multiple systems, and individually may not look suspicious. This creates security gaps attackers can exploit.

Solving the Identity Security Challenge

Identity security is complex and difficult to address. Threats are constant and many, with users and machines targeted with increasingly innovative attack methods by focused cyberattackers. A compromised account can be highly valuable to an attacker, offering hard-to-detect access that can be used to carry out reconnaissance and craft a targeted attack to deploy malware or steal data or funds. The problem of compromised identities is only going to grow, and the impact of compromise is significant, as in many cases, organizations do not have the tools or knowledge to deal with it.

It was the challenge of securing user identities that made me leap at the chance to work on a GigaOm research project into identity threat detection and response (ITDR) solutions, providing me with a chance to learn and understand how security vendors could help address this complex challenge. ITDR solutions are a growing IT industry trend, and while they are a discipline rather than a product, the trend has led to software-based solutions that help enforce that discipline.

How to Choose the Right ITDR Solution

Solution Capabilities

ITDR tools bring together identity-based threat telemetry from many sources, including user directories, identity platforms, cloud platforms, SaaS solutions, and other areas such as endpoints and networks. They then apply analytics, machine learning, and human oversight to look for correlations across data points to provide insight into potential threats.

Critically, they do this quickly and accurately—within minutes—and it is this speed that is essential in tackling threats. In the examples I mentioned, it took days before the identity compromise was spotted, and by then the damage had been done. Tools that can quickly notify of threats and even automate the response will significantly reduce the risk of potential compromise.

Proactive security that can help reduce risk in the first place adds additional value. ITDR solutions can help build a picture of the current environment and apply risk templates to it to highlight areas of concern, such as accounts or data repositories with excessive permissions, unused accounts, and accounts found on the dark web. The security posture insights provided by highlighting these concerns help improve security baselines.

Deception technology is also useful. It works by using fake accounts or resources to attract attackers, leaving the true resources untouched. This reduces the risk to actual resources while providing a useful way to study attacks in progress without risking valuable assets.

Vendor Approach

ITDR solutions fall into two main camps, and while neither approach is better or worse than the other, they are likely to appeal to different markets.

One route is the “add-on” approach, usually from vendors either in the extended detection and response (XDR) space or privileged access management (PAM) space. This approach uses existing insights and applies identity threat intelligence to them. For organizations using XDR or PAM tools already, adding ITDR to can be an attractive option, as they are likely to have more robust and granular mitigation controls and the capability to use other parts of their solution stack to help isolate and stop attacks.

The other approach comes from vendors that have built specific, identity-focused tools from the ground up, designed to integrate broadly with existing technology stacks. These tools pull telemetry from the existing stacks into a dedicated ITDR engine and use that to highlight and prioritize risk and potentially enforce isolation and mitigation. The flexibility and breadth of coverage these tools offer can make them attractive to users with broader and more complex environments that want to add identity security without changing other elements of their current investment.

Next Steps

To learn more, take a look at GigaOm’s ITDR Key Criteria and Radar reports. These reports provide a comprehensive overview of the market, outline the criteria you’ll want to consider in a purchase decision, and evaluate how a number of vendors perform against those decision criteria.

If you’re not yet a GigaOm subscriber, sign up here.

The Challenge of Securing User Identities Read More »

what-kind-of-bug-would-make-machine-learning-suddenly-40%-worse-at-nethack?

What kind of bug would make machine learning suddenly 40% worse at NetHack?

AI, bugs, CUDA, gaming, imitation learning, machine learning, moon phase, nethack, reinforcement learning, roguelike, roguelikes, software bugs / /

Large Moon Models (LMMs) —

One day, a roguelike-playing system just kept biffing it, for celestial reasons.

–

Moon rendered in ASCII text, with

Aurich Lawson

Members of the Legendary Computer Bugs Tribunal, honored guests, if I may have your attention? I would, humbly, submit a new contender for your esteemed judgment. You may or may not find it novel, you may even deign to call it a “bug,” but I assure you, you will find it entertaining.

Consider NetHack. It is one of the all-time roguelike games, and I mean that in the more strict sense of that term. The content is procedurally generated, deaths are permanent, and the only thing you keep from game to game is your skill and knowledge. I do understand that the only thing two roguelike fans can agree on is how wrong the third roguelike fan is in their definition of roguelike, but, please, let us move on.

NetHack is great for machine learning…

Being a difficult game full of consequential choices and random challenges, as well as a “single-agent” game that can be generated and played at lightning speed on modern computers, NetHack is great for those working in machine learning—or imitation learning, actually, as detailed in Jens Tuyls’ paper on how compute scaling affects single-agent game learning. Using Tuyls’ model of expert NetHack behavior, Bartłomiej Cupiał and Maciej Wołczyk trained a neural network to play and improve itself using reinforcement learning.

By mid-May of this year, the two had their model consistently scoring 5,000 points by their own metrics. Then, on one run, the model suddenly got worse, on the order of 40 percent. It scored 3,000 points. Machine learning generally, gradually, goes in one direction with these types of problems. It didn’t make sense.

Cupiał and Wołczyk tried quite a few things: reverting their code, restoring their entire software stack from a Singularity backup, and rolling back their CUDA libraries. The result? 3,000 points. They rebuild everything from scratch, and it’s still 3,000 points.

<em>NetHack</em>, played by a regular human.” height=”506″ src=”https://cdn.arstechnica.net/wp-content/uploads/2024/06/13863751533_64654db44e_o.png” width=”821″></img><figcaption> <p><em>NetHack</em>, played by a regular human.</p> </figcaption></figure> <h2>… except on certain nights</h2> <p>As <a href=detailed in Cupiał’s X (formerly Twitter) thread, this was several hours of confused trial and error by him and Wołczyk. “I am starting to feel like a madman. I can’t even watch a TV show constantly thinking about the bug,” Cupiał wrote. In desperation, he asks model author Tuyls if he knows what could be wrong. He wakes up in Kraków to an answer:

“Oh yes, it’s probably a full moon today.”

In NetHack, the game in which the DevTeam has thought of everything, if the game detects from your system clock that it should be a full moon, it will generate a message: “You are lucky! Full moon tonight.” A full moon imparts a few player benefits: a single point added to Luck, and werecreatures mostly kept to their animal forms.

It’s an easier game, all things considered, so why would the learning agent’s score be lower? It simply doesn’t have data about full moon variables in its training data, so a branching series of decisions likely leads to lesser outcomes, or just confusion. It was indeed a full moon in Kraków when the 3,000-ish scores started showing up. What a terrible night to have a learning model.

Of course, “score” is not a real metric for success in NetHack, as Cupiał himself noted. Ask a model to get the best score, and it will farm the heck out of low-level monsters because it never gets bored. “Finding items required for [ascension] or even [just] doing a quest is too much for pure RL agent,” Cupiał wrote. Another neural network, AutoAscend, does a better job of progressing through the game, but “even it can only solve sokoban and reach mines end,” Cupiał notes.

Is it a bug?

I submit to you that, although NetHack responded to the full moon in its intended way, this quirky, very hard-to-fathom stop on a machine-learning journey was indeed a bug and a worthy one in the pantheon. It’s not a Harvard moth, nor a 500-mile email, but what is?

Because the team used Singularity to back up and restore their stack, they inadvertently carried forward the machine time and resulting bug each time they tried to solve it. The machine’s resulting behavior was so bizarre, and seemingly based on unseen forces, that it drove a coder into fits. And the story has a beginning, a climactic middle, and a denouement that teaches us something, however obscure.

The NetHack Lunar Learning Bug is, I submit, quite worth memorializing. Thank you for your time.

What kind of bug would make machine learning suddenly 40% worse at NetHack? Read More »

isps-seek-halt-of-net-neutrality-rules-before-they-take-effect-next-month

ISPs seek halt of net neutrality rules before they take effect next month

fcc net neutrality, Policy / /

Net neutrality back in court —

Fate of net neutrality may hinge on Supreme Court’s “major questions” doctrine.

–

Illustration of network data represented by curving lines flowing on a dark background.

Getty Images | Yuichiro Chino

As expected, broadband industry lobby groups have sued the Federal Communications Commission in an attempt to nullify net neutrality rules that prohibit blocking, throttling, and paid prioritization.

Lobby groups representing cable, telecom, and mobile Internet service providers sued the FCC in several US appeals courts last week. Industry groups also filed a petition with the FCC on Friday asking for a stay of the rules, claiming the regulations shouldn’t take effect while litigation is pending because the industry is likely to prevail in court.

The FCC is highly likely to reject the petition for a stay, but the groups can then ask appeals court judges to impose an injunction that would prevent enforcement. The industry lost a similar case during the Obama era, but is hoping to win this time because of the Supreme Court’s evolving approach on whether federal agencies can decide “major questions” without explicit instructions from Congress.

The petition for a stay was filed by groups including NCTA-The Internet & Television Association, which represents large cable providers such as Comcast and Charter; and USTelecom, which represents telcos including AT&T, Verizon, and CenturyLink/Lumen.

“By reclassifying broadband under Title II of the Communications Act of 1934, the Commission asserts the power to set prices, dictate terms and conditions, require or prohibit investment or divestment, and more. It should be ‘indisputable’ that the major-questions doctrine applies to that seismic claim of authority,” the petition for a stay said.

Broadband classified as telecommunications

The FCC’s net neutrality order reclassified broadband as telecommunications, which makes Internet service subject to common-carrier regulations under Title II. The order reverses the Trump-era FCC’s classification of broadband as an information service and is scheduled to take effect on July 22. The FCC approved it in a 3-2 vote on April 25.

Despite the industry’s claim that classification is a major question that can only be decided by Congress, a federal appeals court ruled in previous cases that the FCC has authority to classify broadband as either a telecommunications or information service.

The lobby groups claim that without a stay preventing enforcement, their members “will suffer irreparable harm, as they did in the wake of the 2015 Order. In particular, petitioners’ members will be forced to delay or forego valuable new services, incur prohibitive compliance costs, and pay more to obtain capital.”

Lawsuits against the FCC were filed in the US Court of Appeals for the District of Columbia Circuit by CTIA-The Wireless Association, which represents mobile providers; America’s Communications Association (ACA), which represents small and medium-sized cable providers; and the Wireless Internet Service Providers Association (WISPA), which represents fixed wireless providers.

The FCC was sued in other federal circuit appeals courts by the Texas Cable Association, the Ohio Telecom Association, the Ohio Cable Telecommunications Association, the Missouri Internet & Television Association, and Florida Internet & Television Association.

The cases will be consolidated into one court. The DC Circuit appeals court handled challenges to the Obama-era and Trump-era net neutrality decisions, ruling in favor of the FCC both times. Despite the Trump-era repeal, many ISPs still have to follow net neutrality rules because of regulations imposed by California and other states.

FCC: Authority “clear as day”

FCC Commissioner Geoffrey Starks said before the April 25 vote that the FCC’s authority to regulate broadband as a telecommunications service “is clear as day.”

To find otherwise, a court “would need to conclude that ‘this is a major questions case.’ Yet major questions review is reserved for only ‘extraordinary cases’—and this one doesn’t come close,” Starks said. “There’s no ‘unheralded power’ that we’re purporting to discover in the annals of an old, dusty statute—we’ve been classifying communications services one way or the other for decades, and the 1996 [Telecommunications] Act expressly codified our ability to continue that practice.”

If the industry loses at the appeals-court level again, lobby groups would seek review at the Supreme Court. Their hopes depend partly on Justice Brett Kavanaugh, who argued in a 2017 dissent as a circuit court judge that the “net neutrality rule is unlawful and must be vacated” because “Congress did not clearly authorize the FCC to issue the net neutrality rule.”

The CTIA lawsuit against the FCC said, “Given the undisputed fact that broadband Internet is an essential engine of the nation’s economic, social, and political life, the major-questions doctrine requires the FCC to identify clear statutory authority to subject broadband Internet access service to common-carrier regulation. The Order does not and cannot point to such authority. And to the extent there is any statutory ambiguity, the Order’s Title II approach far exceeds the bounds of reasonable interpretation and infringes rights protected by the Constitution.”

ISPs seek halt of net neutrality rules before they take effect next month Read More »

nvidia-emails:-elon-musk-diverting-tesla-gpus-to-his-other-companies

Nvidia emails: Elon Musk diverting Tesla GPUs to his other companies

Cars, elon musk, NVIDIA, Nvidia H100, Tesla, Twitter, X, xAI / /

why not just make cars? —

The Tesla CEO is accused of diverting resources from the company again.

–

A row of server racks

Enlarge / Tesla will have to rely on its Dojo supercomputer for a while longer after CEO Elon Musk diverted 12,000 Nvidia GPU clusters to X instead.

Tesla

Elon Musk is yet again being accused of diverting Tesla resources to his other companies. This time, it’s high-end H100 GPU clusters from Nvidia. CNBC’s Lora Kolodny reports that while Tesla ordered these pricey computers, emails from Nvidia staff show that Musk instead redirected 12,000 GPUs to be delivered to his social media company X.

It’s almost unheard of for a profitable automaker to pivot its business into another sector, but that appears to be the plan at Tesla as Musk continues to say that the electric car company is instead destined to be an AI and robotics firm instead.

Does Tesla make cars or AI?

That explains why Musk told investors in April that Tesla had spent $1 billion on GPUs in the first three months of this year, almost as much as it spent on R&D, despite being desperate for new models to add to what is now an old and very limited product lineup that is suffering rapidly declining sales in the US and China.

Despite increasing federal scrutiny here in the US, Tesla has reduced the price of its controversial “full-self driving” assist, and the automaker is said to be close to rolling out the feature in China. (Questions remain about how many Chinese Teslas would be able to utilize this feature given that a critical chip was left out of 1.2 million cars built there during the chip shortage.)

Perfecting this driver assist would be very valuable to Tesla, which offers FSD as a monthly subscription as an alternative to a one-off payment. The profit margins for subscription software services vastly outstrip the margins Tesla can make selling physical cars, which dropped to just 5.5 percent for Q1 2024. And Tesla says that massive GPU clusters are needed to develop FSD’s software.

Isn’t Tesla desperate for Nvidia GPUs?

Tesla has been developing its own in-house supercomputer for AI, called Dojo. But Musk has previously said that computer could be redundant if Tesla could source more H100s. “If they could deliver us enough GPUs, we might not need Dojo, but they can’t because they’ve got so many customers,” Musk said during a July 2023 investor day.

Which makes his decision to have his other companies jump all the more notable. In December, an internal Nvidia memo seen by CNBC said, “Elon prioritizing X H100 GPU cluster deployment at X versus Tesla by redirecting 12k of shipped H100 GPUs originally slated for Tesla to X instead. In exchange, original X orders of 12k H100 slated for Jan and June to be redirected to Tesla.”

X and the affiliated xAi are developing generative AI products like large language models.

Not the first time

This is not the first time that Musk has been accused of diverting resources (and his time) from publicly held Tesla to his other privately owned enterprises. In December 2022, US Sen. Elizabeth Warren (D-Mass.) wrote to Tesla asking Tesla to explain whether Musk was diverting Tesla resources to X (then called Twitter):

This use of Tesla employees raises obvious questions about whether Mr. Musk is appropriating resources from a publicly traded firm, Tesla, to benefit his own private company, Twitter. This, of course, would violate Mr. Musk’s legal duty of loyalty to Tesla and trigger questions about the Tesla Board’s responsibility to prevent such actions, and may also run afoul other “anti-tunneling rules that aim to prevent corporate insiders from extracting resources from their firms.”

Musk giving time meant (and compensated) for by Tesla to SpaceX, X, and his other ventures was also highlighted as a problem by the plaintiffs in a successful lawsuit to overturn a $56 billion stock compensation package.

And last summer, the US Department of Justice opened an investigation into whether Musk used Tesla resources to build a mansion for the CEO in Texas; the probe has since expanded to cover behavior stretching back to 2017.

These latest accusations of misuse of Tesla resources come at a time when Musk is asking shareholders to reapprove what is now a $46 billion stock compensation plan.

Nvidia emails: Elon Musk diverting Tesla GPUs to his other companies Read More »

intel-details-new-lunar-lake-cpus-that-will-go-up-against-amd,-qualcomm,-and-apple

Intel details new Lunar Lake CPUs that will go up against AMD, Qualcomm, and Apple

copilot+ PC, intel, intel core, lunar lake, meteor lake, NPU, Tech / /

more lakes —

Lunar Lake returns to a more conventional-looking design for Intel.

–

A high-level breakdown of Intel's next-gen Lunar Lake chips, which preserve some of Meteor Lake's changes while reverting others.

Enlarge / A high-level breakdown of Intel’s next-gen Lunar Lake chips, which preserve some of Meteor Lake’s changes while reverting others.

Intel

Given its recent manufacturing troubles, a resurgent AMD, an incursion from Qualcomm, and Apple’s shift from customer to competitor, it’s been a rough few years for Intel’s processors. Computer buyers have more viable options than they have in many years, and in many ways the company’s Meteor Lake architecture was more interesting as a technical achievement than it was as an upgrade for previous-generation Raptor Lake processors.

But even given all of that, Intel still provides the vast majority of PC CPUs—nearly four-fifths of all computer CPUs sold are Intel’s, according to recent analyst estimates from Canalys. The company still casts a long shadow, and what it does still helps set the pace for the rest of the industry.

Enter its next-generation CPU architecture, codenamed Lunar Lake. We’ve known about Lunar Lake for a while—Intel reminded everyone it was coming when Qualcomm upstaged it during Microsoft’s Copilot+ PC reveal—but this month at Computex the company is going into more detail ahead of availability sometime in Q3 of 2024.

Lunar Lake will be Intel’s first processor with a neural processing unit (NPU) that meets Microsoft’s Copilot+ PC requirements. But looking beyond the endless flow of AI news, it also includes upgraded architectures for its P-cores and E-cores, a next-generation GPU architecture, and some packaging changes that simultaneously build on and revert many of the dramatic changes Intel made for Meteor Lake.

Intel didn’t have more information to share on Arrow Lake, the architecture that will bring Meteor Lake’s big changes to socketed desktop motherboards for the first time. But Intel says that Arrow Lake is still on track for release in Q4 of 2024, and it could be announced at Intel’s annual Innovation event in late September.

Building on Meteor Lake

Lunar Lake continues to use a mix of P-cores and E-cores, which allow the chip to handle a mix of low-intensity and high-performance workloads without using more power than necessary.

Enlarge / Lunar Lake continues to use a mix of P-cores and E-cores, which allow the chip to handle a mix of low-intensity and high-performance workloads without using more power than necessary.

Intel

Lunar Lake shares a few things in common with Meteor Lake, including a chiplet-based design that combines multiple silicon dies into one big one with Intel’s Foveros packaging technology. But in some ways Lunar Lake is simpler and less weird than Meteor Lake, with fewer chiplets and a more conventional design.

Meteor Lake’s components were spread across four tiles: a compute tile that was mainly for the CPU cores, a TSMC-manufactured graphics tile for the GPU rendering hardware, an IO tile to handle things like PCI Express and Thunderbolt connectivity, and a grab-bag “SoC” tile with a couple of additional CPU cores, the media encoding and decoding engine, display connectivity, and the NPU.

Lunar Lake only has two functional tiles, plus a small “filler tile” that seems to exist solely so that the Lunar Lake silicon die can be a perfect rectangle once it’s all packaged together. The compute tile combines all of the processor’s P-cores and E-cores, the GPU, the NPU, the display outputs, and the media encoding and decoding engine. And the platform controller tile handles wired and wireless connectivity, including PCIe and USB, Thunderbolt 4, and Wi-Fi 7 and Bluetooth 5.4.

This is essentially the same split that Intel has used for laptop chips for years and years: one chipset die and one die for the CPU, GPU, and everything else. It’s just that now, those two chips are part of the same silicon die, rather than separate dies on the same processor package. In retrospect it seems like some of Meteor Lake’s most noticeable design departures—the division of GPU-related functions among different tiles, the presence of additional CPU cores inside of the SoC tile—were things Intel had to do to work around the fact that another company was actually manufacturing most of the GPU. Given the opportunity, Intel has returned to a more recognizable assemblage of components.

Intel is shifting to on-package RAM for Meteor Lake, something Apple also uses for its M-series chips.

Enlarge / Intel is shifting to on-package RAM for Meteor Lake, something Apple also uses for its M-series chips.

Intel

Another big packaging change is that Intel is integrating RAM into the CPU package for Lunar Lake, rather than having it installed separately on the motherboard. Intel says this uses 40 percent less power, since it shortens the distance data needs to travel. It also saves motherboard space, which can either be used for other components, to make systems smaller, or to make more room for battery. Apple also uses on-package memory for its M-series chips.

Intel says that Lunar Lake chips can include up to 32GB of LPDDR5x memory. The downside is that this on-package memory precludes the usage of separate Compression-Attached Memory Modules, which combine many of the benefits of traditional upgradable DIMM modules and soldered-down laptop memory.

Intel details new Lunar Lake CPUs that will go up against AMD, Qualcomm, and Apple Read More »

for-the-second-time-in-two-years,-amd-blows-up-its-laptop-cpu-numbering-system

For the second time in two years, AMD blows up its laptop CPU numbering system

AMD, intel, qualcomm, Ryzen, ryzen ai, Tech / /

this again —

AMD reverses course on “decoder ring” numbering system for laptop CPUs.

–

AMD's Ryzen 9 AI 300 series is a new chip and a new naming scheme.

Enlarge / AMD’s Ryzen 9 AI 300 series is a new chip and a new naming scheme.

AMD

Less than two years ago, AMD announced that it was overhauling its numbering scheme for laptop processors. Each digit in its four-digit CPU model numbers picked up a new meaning, which, with the help of a detailed reference sheet, promised to inform buyers of exactly what it was they were buying.

One potential issue with this, as we pointed out at the time, was that this allowed AMD to change over the first and most important of those four digits every single year that it decided to re-release a processor, regardless of whether that chip actually included substantive improvements or not. Thus a “Ryzen 7730U” from 2023 would look two generations newer than a Ryzen 5800U from 2021, despite being essentially identical.

AMD is partially correcting this today by abandoning the self-described “decoder ring” naming system and resetting it to something more conventional.

For its new Ryzen AI laptop processors, codenamed “Strix Point,” AMD is still using the same broad Ryzen 3/5/7/9 number to communicate general performance level plus a one- or two-letter suffix to denote general performance and power level (U for ultraportables, HX for higher-performance chips, and so on). A new three-digit processor number will inform buyers of the chip’s generation in the first digit and denote the specific SKU using the last two digits.

AMD is changing how it numbers its laptop CPUs again.

Enlarge / AMD is changing how it numbers its laptop CPUs again.

AMD

In other words, the company is essentially hitting the undo button.

Like Intel, AMD is shifting from four-digit numbers to three digits. The Strix Point processor numbers will start with the 300 series, which AMD says is because this is the third generation of Ryzen laptop processors with a neural processing unit (NPU) included. Current 7040-series and 8040-series processors with NPUs are not being renamed retroactively, and AMD plans to stop using the 7000- and 8000-series numbering for processor introductions going forward.

AMD wouldn’t describe exactly how it would approach CPU model numbers for new products that used older architectures but did say that new processors that didn’t meet the 40+ TOPS requirement for Microsoft’s Copilot+ program would simply use the “Ryzen” name instead of the new “Ryzen AI” branding. That would include older architectures with slower NPUs, like the current 7040 and 8040-series chips.

Desktop CPUs are, once again, totally unaffected by this change. Desktop processors’ four-digit model numbers and alphabetic suffixes generally tell you all you need to know about their underlying architecture; the new Ryzen 9000 desktop CPUs and the Zen 5 architecture were also announced today.

It seems like a lot of work to do to end up basically where we started, especially when the people at AMD who make and market the desktop chips have been getting by just fine with older model numbers for newly released products when appropriate. But to be fair to AMD, there just isn’t a great way to do processor model numbers in a simple and consistent way, at least not given current market realities:

  • PC OEMs that seem to demand or expect “new” product from chipmakers every year, even though chip companies tend to take somewhere between one and three years to release significantly updated designs.
  • The fact that casual and low-end users don’t actually benefit a ton from performance enhancements, keeping older chips viable for longer.
  • Different subsections of the market that must be filled with slightly different chips (consider chips with vPro versus similar chips without it).
  • The need to “bin” chips—that is, disable small parts of a given silicon CPU or GPU die and then sell the results as a lower-end product—to recoup manufacturing costs and minimize waste.

Apple may come the closest to what the “ideal” would probably be—one number for the overarching chip generation (M1, M3, etc.), one word like “Pro” or “Max” to communicate the general performance level, and a straightforward description of the number of CPU and GPU cores included, to leave flexibility for binning chips. But as usual, Apple occupies a unique position: it’s the only company putting its own processors into its own systems, and the company usually only updates a product when there’s something new to put in it, rather than reflexively announcing new models every time another CES or back-to-school season or Windows version rolls around.

In reverting to more traditional model numbers, AMD has at least returned to a system that people who follow CPUs will be broadly familiar with. It’s not perfect, and it leaves plenty of room for ambiguity as the product lineup gets more complicated. But it’s in the same vein as Intel’s rebranding of 13th-gen Core chips, the whole “Intel Processor” thing, or Qualcomm’s unfriendly eight-digit model numbers for its Snapdragon X Plus and Elite chips. AMD’s new nomenclature is a devil, but at least it’s one we know.

For the second time in two years, AMD blows up its laptop CPU numbering system Read More »

amd’s-next-gen-ryzen-9000-desktop-chips-and-the-zen-5-architecture-arrive-in-july

AMD’s next-gen Ryzen 9000 desktop chips and the Zen 5 architecture arrive in July

AMD, Ryzen, ryzen 7000, ryzen 9000, Tech, Zen 4, zen 5, zen 5c / /

ryzen again —

But AMD says AM4 will hang around for budget PCs well into 2025.

–

  • AMD is announcing Ryzen 9000 and Zen 5, the second CPU architecture for its AM5 platform.

    AMD

  • AMD’s Ryzen 9 9950X heads up the new Ryzen 9000 family.

    AMD

  • There are three other variants here, with 12, 8, and 6 Zen 5 CPU cores. The Ryzen 7000 series launched with chips at the same tiers.

    AMD

  • AMD is also announcing a pair of high-end chipsets, though they don’t offer much that’s new; 600-series boards should all support Ryzen 9000 after a BIOS update.

    AMD

  • The Zen 5 CPU architecture powers the Ryzen 9000 series.

    AMD

  • A handful of architectural highlights from Zen 5.

    AMD

  • The performance improvements with Zen 5 are occasionally quite impressive, but on average we’re looking at a 16 percent increase over Zen 4 at the same clock speeds. That’s decent, but not as good as the move from Zen 3 to Zen 4.

    AMD

It’s been almost two years since AMD introduced its Ryzen 7000 series desktop CPUs and the Zen 4 CPU architecture. Today, AMD is announcing the first concrete details about their successors. The Ryzen 9000 CPUs begin shipping in July.

At a high level, the Ryzen 9000 series and Zen 5 architecture offer mostly incremental improvements over Ryzen 7000 (Ryzen 8000 on the desktop is used exclusively for Zen 4-based G-series CPUs with more powerful integrated GPUs). AMD says that Zen 5 is roughly 16 percent faster than Zen 4 at the same clock speeds, depending on the workload—certainly not nothing, and there are some workloads that perform much better. But that number is far short of the 29 percent jump between Zen 3 and Zen 4.

AMD and Intel have both compensated for mild single-core performance improvements in the past by adding more cores, but Ryzen 9000 doesn’t do that. From the 9600X to the 9950X, the chips offer between 6 and 16 full-size Zen 5 cores, the same as every desktop lineup since Zen 2 and the Ryzen 3000 series. De-lidded shots of the processors indicate that they’re still using a total of two or three separate chiplets: one or two CPU chiplets with up to 8 cores each, and a separate I/O die to handle connectivity. The CPU chiplets are manufactured on a TSMC N4 process, an upgrade from the 5nm process used for Ryzen 7000, while the I/O die is still made with a 6nm TSMC process.

Ryzen 9000 has the same layout as the last few generations of Ryzen desktop CPU—two CPU chiplets with up to eight cores each, and an I/O die to handle connectivity.

Enlarge / Ryzen 9000 has the same layout as the last few generations of Ryzen desktop CPU—two CPU chiplets with up to eight cores each, and an I/O die to handle connectivity.

AMD

These chips include no Zen 5c E-cores, as older rumors suggested. Zen 5c is a version of Zen 5 that is optimized to take up less space in a silicon die, at the expense of higher clock speeds; Zen 5c cores are making their debut in the Ryzen AI 300-series laptop chips AMD also announced today. Boosting the number of E-cores has helped Intel match and surpass AMD’s multi-core performance, though Ryzen’s power consumption and efficiency have both outdone Intel’s throughout the 12th-, 13th-, and 14th-generation Core product cycles. Apple also uses a mix of P-cores and E-cores in its  high-end desktop CPU designs.

Ryzen 9000 doesn’t include any kind of neural processing unit (NPU), nor does AMD mention whether the Ryzen 7000’s RDNA 2-based integrated GPU has been upgraded or improved.

AMD is also announcing new X870 and X870E motherboard chipsets to accompany the new processors; as with the X670, the E-series chipset is actually a pair of chipsets on the same motherboard, boosting the number of available USB ports, M.2 slots, and PCIe slots.

The only real improvement here seems to be that all X870-series boards support USB4 and higher EXPO memory overclocking speeds by default. The chipsets also support PCIe 5.0 speeds for the main PCIe slot and M.2 slot, though the X670 chipsets already did this.

The processors’ power requirements aren’t changing, so users with 600-series motherboards ought to be able to use Ryzen 9000 CPUs with little to no performance penalty following a BIOS update.

  • AMD plans to keep the AM4 socket around as a budget platform until at least 2025, according to this slide.

    AMD

  • To that end, it’s announcing a couple more riffs on the old Zen 3-based Ryzen 5000 series, to entice budget builders and upgraders. Pricing hasn’t been announced.

    AMD

Ryzen 9000 doesn’t seem likely to resolve the biggest issues with the AM5 platform, namely the high costs relative to current-gen Intel systems, the high cost relative to AM4-based systems today, and even the high cost relative to AM4-based systems at the same point in the AM4 socket’s lifespan. Motherboards remain more expensive, DDR5 memory remains more expensive, and there are still no AM5 processors available for significantly less than $200.

According to AMD’s own timeline, it plans to keep the AM4 socket around until at least 2025. AM4 is still a surprisingly decent budget platform given that the socket was introduced eight years ago, and AMD does, in fact, continue to trickle out new Ryzen 5000-series CPUs to give buyers and upgrades more options. But it still means that system builders either need to choose between an expensive platform that has a future or a cheaper platform that’s more or less a dead end.

Listing image by AMD

AMD’s next-gen Ryzen 9000 desktop chips and the Zen 5 architecture arrive in July Read More »

the-refreshed-2024-hyundai-elantra-n-remains-a-darn-good-enthusiast-car

The refreshed 2024 Hyundai Elantra N remains a darn good enthusiast car

car review, Cars, Hyundai Elantra N / /
A blue Hyundai Elantra N

Enlarge / The regular Hyundai Elantra is a perfectly fine compact sedan. But once the boffins at Hyundai N got hold of it, they transformed it into something with a lot more character.

Peter Nelson

Few cars are aimed quite at driving enthusiasts like the wholesome sport compact. In terms of everyday usability and fun factor, little can touch them, and luckily, there’s still a good variety of them on the new market. Among the best is the Hyundai Elantra N, which, for the 2024 model year, received a styling and chassis refresh. Pricing starts at $33,245 for three pedals and a manual gearbox, or $35,515 for a dual-clutch eight-speed, and either is a massive value for the performance and fun factor that they offer.

Amply sporty styling, plenty spacious

The 2024 Elantra N’s biggest change is in its face. Where previously it had beady eyes surrounded in a sea of black trim—kind of like the vehicular equivalent of a Belgian Malinois—its headlight, grille, and intake are now more geometric. Looks are subjective, but I’m a fan of the headlights, and the functional inlets improve radiator and brake cooling over the previous fascia.

Elsewhere, it’s pretty much the same angular four-door wearing some trapezoidal accents across its body panels and a pronounced rear spoiler. A new set of forged 19-inch wheels is wrapped in 245/35/19 Michelin Pilot Sport 4S tires—these also shave off 8.25 lbs (3.75 kg) of unsprung weight at each corner, which bodes well for acceleration and handling.

  • The Elantra N is easy to distinguish via its bodykit and rear wing.

    Peter Nelson

  • Hyundai

  • The Elantra N’s seats hold you in place.

    Peter Nelson

  • There’s a grab handle for the passenger.

    Peter Nelson

  • the turbocharged four-cylinder engine has character.

    Peter Nelson

Inside, the Elantra N is spacious, offering great head- and legroom for tall folks up front and much of the same in the back. The front seats are some of the best on the market, offering excellent firmness and very assuring bolsters to keep you held in place when cornering. The steering wheel’s rim is thick and confidence-inspiring, though I wish I could have telescoped it closer to my torso.

Materials quality is good for its price, with substantial soft-touch surfaces where it matters, and the various switchgear and controls are laid out in a very clean manner—as is its infotainment system, which is lag-free. Conveniently, there’s a grab handle carved into the center console for the front-seat passenger. Finally, the N’s spacious interior translates to good overall visibility from a sporty driving position; you don’t sit up high but rather down in it, as any good sport compact ought to be.

The refreshed 2024 Hyundai Elantra N remains a darn good enthusiast car Read More »

daily-telescope:-the-most-distant-galaxy-found-so-far-is-a-total-surprise

Daily Telescope: The most distant galaxy found so far is a total surprise

daily telescope, Science, Space / /

A delightful surprise —

“Its discovery has profound implications.”

–

Behold, the most distant galaxy found to date.

Enlarge / Behold, the most distant galaxy found to date.

NASA, ESA, CSA, STScI et al.

Welcome to the Daily Telescope. There is a little too much darkness in this world and not enough light, a little too much pseudoscience and not enough science. We’ll let other publications offer you a daily horoscope. At Ars Technica, we’re going to take a different route, finding inspiration from very real images of a universe that is filled with stars and wonder.

Good morning. It’s June 1, and today’s photo comes from the James Webb Space Telescope. It’s a banger.

This telescope, launched 18 months ago now, had as one of its express goals to deliver insights about the early Universe. The most straightforward way of doing so is to collect the faintest, most distant light that has spent the longest time traveling to reach Earth.

In some eye-opening new results, the telescope has found and confirmed the discovery of a very bright galaxy that existed just 300 million years after the Big Bang. Based on their observations, astronomers believe the galaxy is 1,600 light-years across and has a mass several hundreds of millions of times the mass of the Sun.

The galaxy may not have the catchiest name—it’s JADES-GS-z14-0, after the JWST Advanced Deep Extragalactic Survey program—but in every other way, it’s a remarkable find.

“All of these observations, together, tell us that JADES-GS-z14-0 is not like the types of galaxies that have been predicted by theoretical models and computer simulations to exist in the very early universe,” the astronomers said. “Its discovery has profound implications for the predicted number of bright galaxies we see in the early universe.”

Source: NASA, ESA, CSA, STScI, Brant Robertson (UC Santa Cruz), Ben Johnson (CfA), Sandro Tacchella (Cambridge), Phill Cargile (CfA)

Do you want to submit a photo for the Daily Telescope? Reach out and say hello.

Daily Telescope: The most distant galaxy found so far is a total surprise Read More »