The A.I Megathread (LLM , GPT , Development)

[bnew]

Baidu delivers new LLMs ERNIE 4.5 and ERNIE X1 undercutting DeepSeek, OpenAI on cost — but they’re not open source (yet)

Baidu has also announced plans to integrate ERNIE 4.5 and ERNIE X1 into its broader ecosystem, including Baidu Search and the Wenxiaoyan app.

venturebeat.com

Baidu delivers new LLMs ERNIE 4.5 and ERNIE X1 undercutting DeepSeek, OpenAI on cost — but they’re not open source (yet)​

Carl Franzen – March 17, 2025 3:48 PM

March 17, 2025 3:48 PM

Credit: VentureBeat made with Midjourney



Over the weekend, Chinese web search giant Baidu announced the launch of two new AI models, ERNIE 4.5 and ERNIE X1 , a multimodal language model and reasoning model, respectively.

Baidu claims they offer state-of-the-art performance on a variety of metrics, besting DeepSeek’s non-reasoning V3 and OpenAI’s GPT-4.5 (how do you like the close name match Baidu chose as well?) on several third-party benchmark tests such as the C-Eval (assessing Chinese LLM performance on knowledge and reasoning across 52 subjects), CMMLU (massive multitask language understanding in Chinese), and GSM8K (math word problems).

It also claims to undercut the cost of both fellow Chinese wunderkind’s DeepSeek’s R1 reasoning model with ERNIE X1 by 50% and US AI juggernaut OpenAI’s GPT-4.5 with ERNIE 4.5 by 99%, respectively.

Yet both have some important limitations, including a lack of open source licensing in the former case (which DeepSeek R1 offers) and a far reduced context compared to the latter (8,000 tokens instead of 128,000, frankly an astonishingly low amount in this age of million-token-plus context windows. Tokens are how a large AI model represents information, with more meaning more information. A 128,000-token window is akin to a 250-page novel).

As X user @claudeglass noted in a post, the small context window makes it perhaps only suitable for customer service chatbots.

Baidu posted on X that it did plan to make the ERNIE 4.5 model family open source on June 30th, 2025.

Baidu has enabled access to the models through its application programming interface (API) and Chinese-language chatbot rival to ChatGPT, known as “ ERNIE Bot ” — it answers questions, generates text, produces creative writing, and interacts conversationally with users — and made ERNIE Bot free to access.

ERNIE 4.5: A new generation of multimodal AI​

ERNIE 4.5 is Baidu’s latest foundation model, designed as a native multimodal system capable of processing and understanding text, images, audio, and video, and is a clear competitor to OpenAI’s GPT-4.5 model released back in February 2025 .

The model has been optimized for better comprehension, generation, reasoning, and memory. Enhancements include improved hallucination prevention, logical reasoning, and coding capabilities.

According to Baidu, ERNIE 4.5 outperforms GPT-4.5 in multiple benchmarks while maintaining a significantly lower cost.

The model’s advancements stem from several key technologies, including FlashMask Dynamic Attention Masking, Heterogeneous Multimodal Mixture-of-Experts, and Self-feedback Enhanced Post-Training.

ERNIE X1 introduces advanced deep-thinking reasoning capabilities, emphasizing understanding, planning, reflection, and evolution.

Unlike standard multimodal AI models, ERNIE X1 is specifically designed for complex reasoning and tool use, enabling it to perform tasks such as advanced search, document-based Q&A, AI-generated image interpretation, code execution, and web page analysis.

The model supports a range of tools, including Baidu’s academic search, business information search, and franchise research tools. Its development is based on Progressive Reinforcement Learning, End-to-End Training integrating Chains of Thought and Action, and a Unified Multi-Faceted Reward System.

Access and API availability​

Users can now access both ERNIE 4.5 and ERNIE X1 via the official ERNIE Bot website.

For enterprise users and developers, ERNIE 4.5 is now available through Baidu AI Cloud’s Qianfan platform via API access. ERNIE X1 is expected to be available soon.

Pricing for API Access:​

• ERNIE 4.5: Input: $0.55 USD per 1 million tokens Output: $2.2 per 1M tokens
• ERNIE X1: Input: $0.28 per 1M tokens Output: $1.1 per 1M tokens

Compare that to:

• GPT-4.5, which has a noticeably astonishingly high price through the OpenAI API of: Input: $75.00 per 1M tokens Output: $150.00 per 1M tokens

• DeepSeek R1 Input: $0.55 per 1M tokens Output: $2.19 per 1M tokens

Baidu has also announced plans to integrate ERNIE 4.5 and ERNIE X1 into its broader ecosystem, including Baidu Search and the Wenxiaoyan app.

Considerations for enterprise decision-makers​

For CIOs, CTOs, IT leaders, and DevOps teams, the launch of ERNIE 4.5 and ERNIE X1 presents both opportunities and considerations:

• Performance vs. Cost – With pricing significantly lower than competing models, organizations evaluating AI solutions may see cost savings by integrating ERNIE models via API. However, further benchmarking and real-world testing may be necessary to assess performance for specific business applications.
• Multimodal and Reasoning Capabilities – The ability to process and understand text, images, audio, and video could be valuable for businesses in industries such as customer support, content generation, legal tech, and finance.
• Tool Integration – ERNIE X1’s ability to work with tools like advanced search, document-based Q&A, and code interpretation could provide automation and efficiency gains in enterprise environments.
• Ecosystem and Localization – As Baidu’s AI models are optimized for Chinese-language processing and regional knowledge, enterprises working in China or targeting Chinese-speaking markets may find ERNIE models more effective than global alternatives.
• Licensing and Data Privacy – While Baidu has indicated that GPT-4.5 will be made open source later this summer, June 30, 2025, that’s still three months away, so enterprises should at least wait until that time to assess whether it’s worth deploying locally or on US-hosted cloud services. Enterprise users should review Baidu’s policies regarding data privacy, compliance, and model usage before integrating these AI solutions.

AI expansion and future outlook​

As AI development accelerates in 2025, Baidu is positioning itself as a leader in multimodal and reasoning-based AI technologies.

The company plans to continue investing in artificial intelligence, data centers, and cloud infrastructure to enhance the capabilities of its foundation models.

By offering a combination of powerful performance and lower costs, Baidu’s latest AI models aim to provide businesses and individual users with more accessible and advanced AI tools.

For more details, visit ERNIE Bot’s official website .

 

[bnew]

Inching towards AGI: How reasoning and deep research are expanding AI from statistical prediction to structured problem-solving

In March 2023, OpenAI released GPT-4, which promised "sparks" of AGI. Two years on, the flame is beginning to appear.

venturebeat.com

Inching towards AGI: How reasoning and deep research are expanding AI from statistical prediction to structured problem-solving​

Gary Grossman, Edelman – March 16, 2025 11:50 AM

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More

AI has evolved at an astonishing pace . What seemed like science fiction just a few years ago is now an undeniable reality. Back in 2017, my firm launched an AI Center of Excellence. AI was certainly getting better at predictive analytics and many machine learning (ML) algorithms were being used for voice recognition, spam detection, spell checking (and other applications) — but it was early. We believed then that we were only in the first inning of the AI game.

The arrival of GPT-3 and especially GPT 3.5 — which was tuned for conversational use and served as the basis for the first ChatGPT in November 2022 — was a dramatic turning point, now forever remembered as the “ChatGPT moment.”

Since then, there has been an explosion of AI capabilities from hundreds of companies. In March 2023 OpenAI released GPT-4, which promised “ sparks of AGI ” (artificial general intelligence). By that time, it was clear that we were well beyond the first inning. Now, it feels like we are in the final stretch of an entirely different sport.

The flame of AGI​

Two years on, the flame of AGI is beginning to appear.

On a recent episode of the Hard Fork podcast , Dario Amodei — who has been in the AI industry for a decade, formerly as VP of research at OpenAI and now as CEO of Anthropic — said there is a 70 to 80% chance that we will have a “very large number of AI systems that are much smarter than humans at almost everything before the end of the decade, and my guess is 2026 or 2027.”

Anthropic CEO Dario Amodei appearing on the Hard Fork podcast. Source:The evidence for this prediction is becoming clearer. Late last summer, OpenAI launched o1 — the first “reasoning model.” They’ve since released o3, and other companies have rolled out their own reasoning models, including Google and, famously, DeepSeek. Reasoners use chain-of-thought (COT), breaking down complex tasks at run time into multiple logical steps, just as a human might approach a complicated task. Sophisticated AI agents including OpenAI’s deep research and Google’s AI co-scientist have recently appeared, portending huge changes to how research will be performed.

Unlike earlier large language models (LLMs) that primarily pattern-matched from training data, reasoning models represent a fundamental shift from statistical prediction to structured problem-solving. This allows AI to tackle novel problems beyond its training, enabling genuine reasoning rather than advanced pattern recognition.

I recently used Deep Research for a project and was reminded of the quote from Arthur C. Clarke: “Any sufficiently advanced technology is indistinguishable from magic.” In five minutes, this AI produced what would have taken me 3 to 4 days. Was it perfect? No. Was it close? Yes, very. These agents are quickly becoming truly magical and transformative and are among the first of many similarly powerful agents that will soon come onto the market.

The most common definition of AGI is a system capable of doing almost any cognitive task a human can do. These early agents of change suggest that Amodei and others who believe we are close to that level of AI sophistication could be correct, and that AGI will be here soon. This reality will lead to a great deal of change, requiring people and processes to adapt in short order.

But is it really AGI?​

There are various scenarios that could emerge from the near-term arrival of powerful AI. It is challenging and frightening that we do not really know how this will go. New York Times columnist Ezra Klein addressed this in a recent podcast : “We are rushing toward AGI without really understanding what that is or what that means.” For example, he claims there is little critical thinking or contingency planning going on around the implications and, for example, what this would truly mean for employment.

Of course, there is another perspective on this uncertain future and lack of planning, as exemplified by Gary Marcus, who believes deep learning generally (and LLMs specifically) will not lead to AGI. Marcus issued what amounts to a take down of Klein’s position, citing notable shortcomings in current AI technology and suggesting it is just as likely that we are a long way from AGI.

Marcus may be correct, but this might also be simply an academic dispute about semantics. As an alternative to the AGI term, Amodei simply refers to “powerful AI” in his Machines of Loving Grace blog , as it conveys a similar idea without the imprecise definition, “sci-fi baggage and hype.” Call it what you will, but AI is only going to grow more powerful.

Playing with fire: The possible AI futures​

In a 60 Minutesinterview , Alphabet CEO Sundar Pichai said he thought of AI as “the most profound technology humanity is working on. More profound than fire, electricity or anything that we have done in the past.” That certainly fits with the growing intensity of AI discussions. Fire, like AI, was a world-changing discovery that fueled progress but demanded control to prevent catastrophe. The same delicate balance applies to AI today.

A discovery of immense power, fire transformed civilization by enabling warmth, cooking, metallurgy and industry. But it also brought destruction when uncontrolled. Whether AI becomes our greatest ally or our undoing will depend on how well we manage its flames. To take this metaphor further, there are various scenarios that could soon emerge from even more powerful AI:

• The controlled flame (utopia): In this scenario, AI is harnessed as a force for human prosperity. Productivity skyrockets, new materials are discovered, personalized medicine becomes available for all, goods and services become abundant and inexpensive and individuals are freed from drudgery to pursue more meaningful work and activities. This is the scenario championed by many accelerationists, in which AI brings progress without engulfing us in too much chaos.
• The unstable fire (challenging): Here, AI brings undeniable benefits — revolutionizing research, automation, new capabilities, products and problem-solving. Yet these benefits are unevenly distributed — while some thrive, others face displacement, widening economic divides and stressing social systems. Misinformation spreads and security risks mount. In this scenario, society struggles to balance promise and peril. It could be argued that this description is close to present-day reality.
• The wildfire (dystopia): The third path is one of disaster, the possibility most strongly associated with so-called “doomers” and “probability of doom” assessments. Whether through unintended consequences, reckless deployment or AI systems running beyond human control, AI actions become unchecked, and accidents happen. Trust in truth erodes. In the worst-case scenario, AI spirals out of control, threatening lives, industries and entire institutions.

While each of these scenarios appears plausible, it is discomforting that we really do not know which are the most likely, especially since the timeline could be short. We can see early signs of each: AI-driven automation increasing productivity, misinformation that spreads at scale, eroding trust and concerns over disingenuous models that resist their guardrails. Each scenario would cause its own adaptations for individuals, businesses, governments and society.

Our lack of clarity on the trajectory for AI impact suggests that some mix of all three futures is inevitable. The rise of AI will lead to a paradox, fueling prosperity while bringing unintended consequences. Amazing breakthroughs will occur, as will accidents. Some new fields will appear with tantalizing possibilities and job prospects, while other stalwarts of the economy will fade into bankruptcy.

We may not have all the answers, but the future of powerful AI and its impact on humanity is being written now. What we saw at the recent Paris AI Action Summit was a mindset of hoping for the best, which is not a smart strategy. Governments, businesses and individuals must shape AI’s trajectory before it shapes us. The future of AI won’t be determined by technology alone, but by the collective choices we make about how to deploy it.

Gary Grossman is EVP of technology practice at Edelman.

 

[bnew]

Cloudflare turns AI against itself with endless maze of irrelevant facts

New approach punishes AI companies that ignore “no crawl” directives.

arstechnica.com

Cloudflare turns AI against itself with endless maze of irrelevant facts​

New approach punishes AI companies that ignore "no crawl" directives.

Benj Edwards – Mar 21, 2025 5:14 PM | 77




Credit: iambuff via Getty Images

On Wednesday, web infrastructure provider Cloudflare announced a new feature called " AI Labyrinth " that aims to combat unauthorized AI data scraping by serving fake AI-generated content to bots. The tool will attempt to thwart AI companies that crawl websites without permission to collect training data for large language models that power AI assistants like ChatGPT .

Cloudflare, founded in 2009, is probably best known as a company that provides infrastructure and security services for websites, particularly protection against distributed denial-of-service (DDoS) attacks and other malicious traffic.

Instead of simply blocking bots, Cloudflare's new system lures them into a "maze" of realistic-looking but irrelevant pages, wasting the crawler's computing resources. The approach is a notable shift from the standard block-and-defend strategy used by most website protection services. Cloudflare says blocking bots sometimes backfires because it alerts the crawler's operators that they've been detected.

"When we detect unauthorized crawling, rather than blocking the request, we will link to a series of AI-generated pages that are convincing enough to entice a crawler to traverse them," writes Cloudflare. "But while real looking, this content is not actually the content of the site we are protecting, so the crawler wastes time and resources."

The company says the content served to bots is deliberately irrelevant to the website being crawled, but it is carefully sourced or generated using real scientific facts—such as neutral information about biology, physics, or mathematics—to avoid spreading misinformation (whether this approach effectively prevents misinformation, however, remains unproven). Cloudflare creates this content using its Workers AI service, a commercial platform that runs AI tasks.

Cloudflare designed the trap pages and links to remain invisible and inaccessible to regular visitors, so people browsing the web don't run into them by accident.

A smarter honeypot​

AI Labyrinth functions as what Cloudflare calls a "next-generation honeypot." Traditional honeypots are invisible links that human visitors can't see but bots parsing HTML code might follow. But Cloudflare says modern bots have become adept at spotting these simple traps, necessitating more sophisticated deception. The false links contain appropriate meta directives to prevent search engine indexing while remaining attractive to data-scraping bots.

"No real human would go four links deep into a maze of AI-generated nonsense," Cloudflare explains. "Any visitor that does is very likely to be a bot, so this gives us a brand-new tool to identify and fingerprint bad bots."

This identification feeds into a machine learning feedback loop—data gathered from AI Labyrinth is used to continuously enhance bot detection across Cloudflare's network, improving customer protection over time. Customers on any Cloudflare plan—even the free tier—can enable the feature with a single toggle in their dashboard settings.

A growing problem​

Cloudflare's AI Labyrinth joins a growing field of tools designed to counter aggressive AI web crawling. In January, we reported on "Nepenthes," software that similarly lures AI crawlers into mazes of fake content. Both approaches share the core concept of wasting crawler resources rather than simply blocking them. However, while Nepenthes' anonymous creator described it as "aggressive malware" meant to trap bots for months, Cloudflare positions its tool as a legitimate security feature that can be enabled easily on its commercial service.

The scale of AI crawling on the web appears substantial, according to Cloudflare's data that lines up with anecdotal reports we've heard from sources. The company says that AI crawlers generate more than 50 billion requests to their network daily, amounting to nearly 1 percent of all web traffic they process. Many of these crawlers collect website data to train large language models without permission from site owners, a practice that has sparked numerous lawsuits from content creators and publishers.

The technique represents an interesting defensive application of AI, protecting website owners and creators rather than threatening their intellectual property. However, it's unclear how quickly AI crawlers might adapt to detect and avoid such traps, potentially forcing Cloudflare to increase the complexity of its deception tactics. Also, wasting AI company resources might not please people who are critical of the perceived energy and environmental costs of running AI models.

Cloudflare describes this as just "the first iteration" of using AI defensively against bots. Future plans include making the fake content harder to detect and integrating the fake pages more seamlessly into website structures. The cat-and-mouse game between websites and data scrapers continues, with AI now being used on both sides of the battle.

 

[bnew]

Nvidia adopts Google’s SynthID watermarking.

In an effort to help preserve intellectual property and fight misinformation and misattribution, AI-generated video outputs from Nvidia’s Cosmos world foundation models will include SynthID digital watermarks. That makes Nvidia the first external user of Google DeepMind’s watermarking tech...

www.theverge.com

Nvidia adopts Google’s SynthID watermarking. | The Verge​

Posted Mar 18, 2025 at 4:00 PM EDT

Nvidia adopts Google’s SynthID watermarking.

In an effort to help preserve intellectual property and fight misinformation and misattribution, AI-generated video outputs from Nvidia’s Cosmos world foundation models will include SynthID digital watermarks. That makes Nvidia the first external user of Google DeepMind’s watermarking tech, announced back in 2023.

Update: Nvidia now says only video outputs will be watermarked, not text and audio as originally told to The Verge.

 

[bnew]

xAI acquires text-to-video startup Hotshot.

Aakash Sastry, the founder of Hotshot, said in a post on X that the AI startup is “excited to continue scaling” its efforts at the Elon Musk-owned xAI. This could be a sign that xAI is getting closer to adding video generation to its AI chatbot Grok, something Musk previously hinted at in...

www.theverge.com

xAI acquires text-to-video startup Hotshot. | The Verge​

Posted Mar 17, 2025 at 4:04 PM EDT

xAI acquires text-to-video startup Hotshot.

Aakash Sastry, the founder of Hotshot, said in a post on X that the AI startup is “excited to continue scaling” its efforts at the Elon Musk-owned xAI. This could be a sign that xAI is getting closer to adding video generation to its AI chatbot Grok, something Musk previously hinted at in January, as spotted by TechCrunch.

 

[bnew]

Hundreds of celebrities warn against letting OpenAI and Google ‘freely exploit’ Hollywood

Celebrities want to uphold existing copyright laws.

www.theverge.com

Hundreds of celebrities warn against letting OpenAI and Google ‘freely exploit’ Hollywood​

Ben Stiller, Mark Ruffalo, Aubrey Plaza, and others say OpenAI and Google want an exemption to train AI on copyrighted materials.

by Emma Roth

Mar 18, 2025, 9:42 AM EDT

4 Comments

Photo by Erika Goldring/Getty Images

Emma Roth is a news writer who covers the streaming wars, consumer tech, crypto, social media, and much more. Previously, she was a writer and editor at MUO.

More than 400 members of the entertainment industry have signed a letter pushing back on OpenAI and Google’s proposal to allow AI models to train on copyrighted content, as reported earlier by Variety. The letter claims both companies “are arguing for a special government exemption” that would allow them to “freely exploit” creative industries.

The letter, which comes in response to the Trump administration’s request for feedback on its incoming AI Action Plan, is signed by stars like Ben Stiller, Mark Ruffalo, Cynthia Eviro, Cate Blanchett, Taika Waititi, Ayo Edebiri, Aubrey Plaza, Guillermo del Toro, Natasha Lyonne, Paul McCartney, and many others.

It directly addresses Google and OpenAI’s comments on the AI Action Plan, both of which argue that they need access to copyrighted materials to train their AI models and that existing laws are holding them back. In its letter, OpenAI claims applying fair use protections to AI is a “matter of national security.”

“There is no reason to weaken or eliminate the copyright protections that have helped America flourish,” the celebrities’ letter reads. “Not when AI companies can use our copyrighted material by simply doing what the law requires: negotiating appropriate licenses with copyright holders — just as every other industry does.”

In a statement to The Verge, Google spokesperson Michael Appel said the company supports “America’s existing fair use framework, and we’re confident that current copyright law enables AI innovation.” Last year, Hollywood stars banded together in support of California’s AI safety bill, which was later vetoed by Governor Gavin Newsom.

You can read the letter in its entirety below:

Update, March 19th: Added a statement from Google.

 

[bnew]

OpenAI and Google ask the government to let them train AI on content they don’t own

OpenAI says it’s a “matter of national security.”

www.theverge.com

OpenAI and Google ask the government to let them train AI on content they don’t own​

OpenAI argues it needs access to ‘avoid forfeiting’ the lead in AI to China.

by Emma Roth

Mar 14, 2025, 3:44 PM EDT

0 Comments

Image: Hugo Herrera / The Verge

Emma Roth is a news writer who covers the streaming wars, consumer tech, crypto, social media, and much more. Previously, she was a writer and editor at MUO.

OpenAI and Google are pushing the US government to allow their AI models to train on copyrighted material. Both companies outlined their stances in proposals published this week, with OpenAI arguing that applying fair use protections to AI “is a matter of national security.”

The proposals come in response to a request from the White House, which asked governments, industry groups, private sector organizations, and others for input on President Donald Trump’s “AI Action Plan.” The initiative is supposed to “enhance America’s position as an AI powerhouse,” while preventing “burdensome requirements” from impacting innovation.

In its comment, Open claims that allowing AI companies to access copyrighted content would help the US “avoid forfeiting” its lead in AI to China, while calling out the rise of DeepSeek.

“There’s little doubt that the PRC’s [People’s Republic of China] AI developers will enjoy unfettered access to data — including copyrighted data — that will improve their models,” OpenAI writes. “If the PRC’s developers have unfettered access to data and American companies are left without fair use access, the race for AI is effectively over.”

Google, unsurprisingly, agrees. The company’s response similarly states that copyright, privacy, and patents policies “can impede appropriate access to data necessary for training leading models.” It adds that fair use policies, along with text and data mining exceptions, have been “critical” to training AI on publicly available data.

“These exceptions allow for the use of copyrighted, publicly available material for AI training without significantly impacting rightsholders and avoid often highly unpredictable, imbalanced, and lengthy negotiations with data holders during model development or scientific experimentation,” Google says.

Anthropic, the AI company behind the AI chatbot Claude, also submitted a proposal – but it doesn’t mention anything about copyrights. Instead, it asks the US government to develop a system to assess an AI model’s national security risks and to strengthen export controls on AI chips. Like Google and OpenAI, Anthropic also suggests that the US bolster its energy infrastructure to support the growth of AI.

Many AI companies have been accused of ripping copyrighted content to train their AI models. OpenAI currently faces several lawsuits from news outlets, including The New York Times, and has even been sued by well-known names like Sarah Silverman and George R.R. Martin. Apple, Anthropic, and Nvidia have also been accused of scraping YouTube subtitles to train AI, which YouTube has said violates its terms.

 

[bnew]

Zoom’s AI will soon help you do busywork, too

Even Zoom is getting in on the agentic AI craze.

www.theverge.com

Zoom’s AI will soon help you do busywork, too​

Zoom’s AI Companion will be able to schedule meetings for you.

by Emma Roth

Mar 17, 2025, 9:21 AM EDT

1 Comments

Illustration by Alex Castro / The Verge

Emma Roth is a news writer who covers the streaming wars, consumer tech, crypto, social media, and much more. Previously, she was a writer and editor at MUO.

Zoom’s AI Companion is getting an “agentic” upgrade at the end of this month, allowing it to identify and perform tasks on users’ behalf. Workplace users will be able to access the feature within a new Tasks tab in Workplace, where they can have Zoom’s AI schedule follow-up meetings, generate documents from meetings, and create video clips.

Some other updates are coming to Zoom too, including a new voice recorder for the Zoom Workplace mobile app that will record, transcribe, and summarize in-person meetings when it launches this month. Zoom is also rolling out live notes in May, which will generate real-time summaries during Zoom meetings and phone calls. The app’s AI Companion and these new features will be available at no additional cost to Zoom Workplace users.

“We see AI stretching across every part of every product and interactions in ways that is really helpful and useful so that it can help you do more,” Smita Hashim, Zoom’s chief product officer, said during an interview with The Verge. “Think of our AI companion. It’s touching every product at this point, but it’s also bringing all of the products together in a way that users can get the work done right.”

Next month, Zoom also plans to launch its $12 / month Custom AI Companion add-on. This includes access to Zoom’s custom avatar feature, which generates an AI version of yourself that you can use to send messages to your team. It’s also adding generic avatar templates to Zoom at no added cost.

 

[bnew]

Roblox’s new AI model can generate 3D objects

Eventually, it will be multimodal.

www.theverge.com

Roblox’s new AI model can generate 3D objects​

The model, Cube 3D, creates 3D models from a text prompt.

by Jay Peters

Mar 17, 2025, 9:30 AM EDT

1 Comments

Illustration: The Verge

Jay Peters is a news editor covering technology, gaming, and more. He joined The Verge in 2019 after nearly two years at Techmeme.

Roblox is launching and open-sourcing Cube 3D, the first version of its foundational AI model for generating 3D objects.

“With Cube, we intend to make 3D creation more efficient,” Roblox says in a press release. “With 3D mesh generation, developers can quickly explore new creative directions and increase their productivity by deciding rapidly which to move forward with.”

The model is trained on 3D objects, as Roblox explains in its press release:

To achieve this, we’ve taken inspiration from state-of-the-art models trained on text tokens (or sets of characters) so they can predict the next token to form a sentence. Our innovation builds on the same core idea. We’ve built the ability to tokenize 3D objects and understand shapes as tokens and trained Cube 3D to predict the next shape token to build a complete 3D object.

The actual data used to train the model includes “a combination of licensed and publicly available datasets, as well as experience data from the Roblox ecosystem,” spokesperson Samantha Spielman tells The Verge.

Down the line, Cube 3D will be able to generate objects using images as inputs, too. “It will ultimately be a multimodal model, trained on text, images, video, and other types of input – and will integrate with our existing AI creation tools,” according to Roblox.

 

[bnew]

Chip race: Microsoft, Meta, Google, and Nvidia battle it out for AI chip supremacy

Powerful chips are powering AI, and companies want to take advantage.

www.theverge.com

Intel is reportedly testing its 18A process again. | The Verge​

Posted Mar 18, 2025, 2:49 PM EDT

Intel is reportedly testing its 18A process again.

After a test of its 18A process last year reportedly failed, Reuters says both Nvidia and Broadcom are actively testing it. The 18A process is a key to Intel’s plan to reestablish itself in the race to build new AI chips.



Nvidia is hosting on March 18th, and its keynote speaker has just revealed his talk. “Come to GTC and I’ll talk to you about Blackwell Ultra, Vera Rubin, and then show you the one click after that,” Nvidia CEO Jensen Huang told analysts on the fiscal Q4 2025 earnings call.

He says Blackwell Ultra will come in the second half of next year, with new networking, new memory, new processors, but on the same system architecture as Blackwell.

“I am not happy with where we are today,” Intel interim co-CEO Michelle Johnston Holthaus says of the company’s AI server business on today’s earnings call, a business which has not met its promises with Gaudi.

So, says Holthaus, Intel will “simplify our roadmap and concentrate our resources” by canceling Falcon Shores. We plan to leverage Falcon Shores as an internal test chip only without bringing it to market.” It’ll focus on Jaguar Shores, a “system-level solution at rack scale,” instead.



As Chinese AI startup DeepSeek draws attention for open-source AI models that it says are cheaper than the competition while providing similar or better performance, AI chip king Nvidia’s stock price dropped today.

CNBC said that after closing at $118.58, down 17 percent, this was “the biggest drop ever for a U.S. company.”

This is Nvidia CEO Jensen Huang’s introduction of Project Digits, a GB10 Grace Blackwell Superchip-powered system with 128GB of RAM that costs about $3,000 and can run sophisticated AI models in a package small enough to sit on your desk.

$14.8 billion profit in Q1, $16.6 billion in Q2, and now $19.3 billion in Q3 of fiscal 2025 — that’s profit, not earnings. (Earnings were $35.08 billion, up from $30.04 billion last quarter.)

The vast majority is from AI data center, of course — but gaming did have a 14 percent bump. It’s a $3B-a-quarter business, while data center is a $30B one.

“It was functional, but the design flaw caused the yield to be low. It was 100% Nvidia’s fault,” Nvidia’s Jensen Huang tells Reuters, effectively confirming The Information’s report from August about why its new flagship AI chips won’t ship in large amounts right away.

He says it’s now fixed, but the timeline stays the same: Q4 for first shipments.

AMD says the MI325X, shipping Q4, will beat Nvidia’s H200. But Nvidia seems a step ahead; it’ll ship “several billion dollars” of its next-gen Blackwell B200 GPU in Q4, too. AMD says its Blackwell competitor, the MI355X, won’t arrive till 2H 2025.

AMD isn’t talking price, but told us it’ll undercut Nvidia when it comes to total cost of ownership.

We’ve heard rumors about AI chip manufacturing projects in the Middle East linked to OpenAI Sam Altman and Elon Musk.

Now, the WSJ says Samsung and TSMC execs have visited the United Arab Emirates “recently,” discussing projects worth up to $100 billion despite concerns about water sources and building up local engineering talent.

Former Bloomberg reporter Tim Culpan writes on Substack:

They’re only used in the iPhone 14 Pro and standard iPhone 15 right now, but maybe the American-made chips Apple signed up for will end up in a future iPhone SE someday. The question is if it’s worth the costs.

 

[bnew]

Nvidia announces Blackwell Ultra GB300 and Vera Rubin, its next AI ‘superchips’

It’s the “tock” in Nvidia’s new “tick-tock” product roadmap.

www.theverge.com

Nvidia announces Blackwell Ultra GB300 and Vera Rubin, its next AI ‘superchips’​

Blackwell Ultra is an enhanced version of Blackwell, but Nvidia’s mostly comparing it to 2022’s H100.

by Sean Hollister

Mar 18, 2025, 2:49 PM EDT

1 Comments

Image: Nvidia

Sean Hollister is a senior editor and founding member of The Verge who covers gadgets, games, and toys. He spent 15 years editing the likes of CNET, Gizmodo, and Engadget.

Nvidia now makes $2,300 in profit every second on the back of the AI revolution. Its data center business is so gigantic, even its networking hardware now rakes in more money than its gaming GPUs. Now, the company is announcing the AI GPUs that it hopes will extend its commanding lead: the Blackwell Ultra GB300, which will ship in the second half of this year, the Vera Rubin for the second half of next year, and the Rubin Ultra that will arrive in the second half of 2027.

This year’s Blackwell Ultra isn’t what we originally expected when Nvidia said last year that it would begin producing new AI chips on a yearly cadence, faster than ever before, as Blackwell Ultra is not on a new architecture. But Nvidia quickly moved on from Blackwell Ultra during today’s GTC keynote to reveal that next architecture, Vera Rubin, whose full rack should offer 3.3x the performance of a comparable Blackwell Ultra one.

Nvidia isn’t making it easy to tell how much better Blackwell Ultra is than the original Blackwell. In a prebriefing with journalists, Nvidia revealed a single Ultra chip will offer the same 20 petaflops of AI performance as Blackwell, but now with 288GB of HBM3e memory rather than 192GB of the same. Meanwhile, a Blackwell Ultra DGX GB300 “Superpod” cluster will offer the same 288 CPUs, 576 GPUs and 11.5 exaflops of FP4 computing as the Blackwell version, but with 300TB of memory rather than 240TB.

Mostly, Nvidia compared its new Blackwell Ultra to the H100, the 2022 chip that originally built Nvidia’s AI fortunes and from which leading companies might presumably want to upgrade: there, Nvidia says it offers 1.5x the FP4 inference and can dramatically speed up “AI reasoning,” with the NVL72 cluster capable of running an interactive copy of DeepSeek-R1 671B that can provide answers in just 10 seconds instead of the H100’s 1.5 minutes. Nvidia says that’s because it can process 1,000 tokens per second, 10 times that of Nvidia’s 2022 chips.

But one intriguing difference is that some companies will be able to buy a single Blackwell Ultra chip: Nvidia announced a desktop computer called the DGX Station with a single GB300 Blackwell Ultra on board, 784GB of unified system memory, built-in 800Gbps Nvidia networking, and the promised 20 petaflops of AI performance. Asus, Dell, and HP will join Boxx, Lambda, and Supermicro in selling versions of the desktop.

Nvidia will also offer a single rack called the GB300 NVL72 that offers 1.1 exaflops of FP4, 20TB of HBM memory, 40TB of “fast memory,” and 130TB/sec of NVLink bandwidth and 14.4 TB/sec networking.

But Vera Rubin and Rubin Ultra may dramatically improve on that performance when they arrive in 2026 and 2027. Rubin has 50 petaflops of FP4, up from 20 petaflops in Blackwell. Rubin Ultra will feature a chip that effectively contains two Rubin GPUs connected together, with twice the performance at 100 petaflops of FP4 and nearly quadruple the memory at 1TB.

A full NVL576 rack of Rubin Ultra claims to offer 15 exaflops of FP4 inference and 5 exaflops of FP8 training for what Nvidia says is 14x the performance of the Blackwell Ultra rack it’s shipping this year. Find other specs by blowing up the images below:

Nvidia says it has already shipped $11 billion worth of Blackwell revenue; the top four buyers alone have purchased 1.8 million Blackwell chips so far in 2025.

Nvidia’s pushing these new chips — and all its AI chips — as essential to the future of computing and is trying to argue today that companies will need more and more computing power, not less as some assumed after DeepSeek shook up investor assumptions and sent Nvidia’s stock price tumbling. At the Nvidia GPU technology conference today, founder and CEO Jensen Huang says the industry needs “100 times more than then we thought we needed this time last year” to keep up with demand.

Huang says Nvidia’s next architecture after Vera Rubin, coming 2028, will be named Feynman — presumably after Richard Feynman, the famous theoretical physicist. He said some of pioneering astronomer Vera Rubin’s family was in the audience today.

 

[bnew]

Amazon is blundering into an AI copyright nightmare

And it may not even realize it.

www.theverge.com

Amazon is blundering into an AI copyright nightmare​

Alexa’s integration with Suno misses the whole point of music.

by Elizabeth Lopatto

Mar 19, 2025, 9:30 AM EDT

0 Comments

Image: Cath Virginia / The Verge

Elizabeth Lopatto is a reporter who writes about tech, money, and human behavior. She joined The Verge in 2014 as science editor. Previously, she was a reporter at Bloomberg.

Suno wasn’t supposed to be an important part of Amazon’s Alexa Plus presentation. The AI song generation platform was a minor demonstration of how Alexa Plus could integrate into other apps, sandwiched between other announcements. But it caught my attention all the same — because whether Amazon realized it or not, the company blundered into a massive copyright fight.

Suno, for those of you not familiar, is an AI song generator: enter a text prompt (such as “a jazz, reggae, EDM pop song about my imagination”) and a song comes back. Like many generative AI companies, it is also being sued by all and sundry for ingesting copyrighted material. The parties in the suit — including major labels and the RIAA — don’t have a smoking gun, since they can’t directly peek at Suno’s training data. But they have managed to generate some suspiciously similar-sounding AI generated materials, mimicking (among others) “Johnny B. Goode,” “Great Balls of Fire,” and Jason Derulo’s habit of singing his own name.

Suno essentially admits these songs were regurgitated from copyrighted source material, but it says such use was legal. “It is no secret that the tens of millions of recordings that Suno’s model was trained on presumably included recordings whose rights are owned by the Plaintiffs in this case,” it says in its own legal filing. Whether AI training data constitutes fair use is a common but unsettled legal argument, and the plaintiffs contend Suno still amounts to “pervasive illegal copying” of artists’ works.

Amazon’s Suno integration, as demonstrated, requires a Suno account to be linked to Alexa. Suno is meant to be hyper-personalized music, letting anyone generate a song. One of the current core features in Alexa’s Echo speakers is taking verbal requests for (non-AI-generated) music. With the Suno demo, Amazon risks antagonizing the very players that make this possible, while simultaneously undercutting Suno’s legal case. Amazon declined to provide an on-the-record comment about the Suno demonstration.

One of the key questions in a fair use lawsuit — including the RIAA’s suit against Suno — is whether a derivative work is intended to replace the original thing. In 2023, the Supreme Court found that Andy Warhol had infringed on photographer Lynn Goldsmith’s copyright when he screenprinted one of her pictures of Prince; a deciding factor was that outlets like Vanity Fair had licensed Warhol’s work instead of Goldsmith’s, offering her no credit or payment.

Every minute spent listening to Suno’s ‘All I Want for Christmas Is You’ is one spent not listening to Mariah Carey’s

This hasn’t been tested with AI music, but the RIAA is making similar arguments, and Amazon’s integration seems to provide a concrete example. Every minute spent listening to Suno’s “All I Want for Christmas Is You” is one spent not listening to Mariah Carey’s “All I Want for Christmas is You” through Spotify or Amazon Music Unlimited — and Carey et al. get stiffed to boot.

If “Suno is suddenly available to every Alexa subscriber, that would be of great concern,” says Richard James Burgess, the president and CEO of the American Association of Independent Music. (Currently, the feature requires both a Suno subscription and either a subscription for Prime or Alexa Plus.) Burgess emphasized that the problem is the alleged copyright violations, not AI-generated music as a whole. “If it hasn’t been licensed correctly from rights holders, then that’s problematic for all music,” he says. “It affects people’s businesses. It affects their livelihoods.”

Suno, like a lot of other AI companies, offers subscriptions that allow users to generate songs, which are not very good. (The free tier allows 10 songs per day.) I’ve seen little about how Suno plans to make a sustainable business, but I do know this: if the company is found to have infringed on copyright, the damages for the songs it’s already used will be sky-high, on top of any other licensing fees Suno will have to pay. That could result in bankruptcy.

I’m not convinced Suno understands why people care about music or what the point is. In a Rolling Stone interview, its cofounder Mikey Shulman complains that musicians are outnumbered by their audience — it’s “so lopsided.” I emailed Shulman to see if he wanted to chat for this article. He didn’t reply.

Music, like all worthwhile art, is about people. If more people want to make music, they can — by learning how to play an instrument or sing. One of the benefits of learning an instrument is that it deepens your appreciation; suddenly you can hear a song’s time signature or notice the difference in feel between keys. You don’t even have to be very good to make music people enjoy — that’s why God created punk!

The AI songs that have broken through to public consciousness have been ones like “BBL Drizzy” and “10 Drunk Cigarettes,” which are not purely AI generated. Rather, there’s a musician working with the AI as a tool to curate and edit it. But that’s not what the Suno demo showed. Instead, it’s just raw prompt generation. This is the least interesting way to interact with generative AI music, and the one that most threatens the actual music industry. An Alexa speaker is not a tool for editing or playing with generative music.

There’s another way in which Suno can undercut real musicians, besides just stealing listening time. The music industry already has a problem with soundalikes and AI-based fraud; Suno’s slop makes it even easier to generate fraudulent tracks.

And Amazon is doing itself no favors here, either. Amazon Music has its own deals with record labels, including the ones suing Suno. In a December 2024 press release, Universal Music Group touts an “expanded global relationship” with Amazon that means the “advancement of artist-centric principles.” It goes on to say that “UMG and Amazon will also work collaboratively to address, among other things, unlawful AI-generated content, as well as protecting against fraud and misattribution.”

So much for that, huh?

 

[bnew]

What does OpenAI really want from Trump?

Strikingly absent in OpenAI’s proposal to Trump is any concept of how AI should be governed — or whether it should be governed at all.

www.theverge.com

What does OpenAI really want from Trump?​

Freedom from state laws — especially SB 1047.

by Tina Nguyen

Mar 19, 2025, 11:15 AM EDT

1 Comments

Image: Abdullah Guclu via Getty Images

Tina Nguyen is a senior reporter for The Verge, covering the Trump administration, Elon Musk’s takeover of the federal government, and the tech industry’s embrace of the MAGA movement.

When AI giant OpenAI submitted its “freedom-focused” policy proposal to the White House’s AI Action Plan last Thursday, it gave the Trump administration an industry wishlist: use trade laws to export American AI dominance against the looming threat of China, loosen copyright restrictions for training data (also to fight China), invest untold billions in AI infrastructure (again: China), and stop states from smothering it with hundreds of new laws.

But specifically, one law: SB 1047, California’s sweeping, controversial, and for now, defeated AI safety bill.

Hundreds of AI-related bills are flooding state governments nationwide, and hundreds or even thousands are expected by the end of 2025, a regulatory deluge the AI industry is trying to dodge. Broad safety regulations like SB 1047 are particularly threatening, posing a perhaps existential threat to OpenAI. Strikingly absent, however, is any notion of how AI should be governed — or whether it should be governed at all.

“OpenAI’s proposal for preemption, just to be totally candid with you, it confuses me a little bit, as someone who’s thought about this issue a lot,” Dean Ball of the libertarian Mercatus Center told me. While federal preemption — the legal concept that federal laws supersede contradicting state laws — is a longstanding practice in several fields, OpenAI’s proposal notably does not include a national framework to replace any AI laws the states are working on. OpenAI did not respond to a request for comment on the plan.

OpenAI’s argument against state regulation reflects a common perception that 50 states enforcing 50 versions of AI law would create chaos for citizens and AI companies alike.

 

[bnew]

Meta AI is rolling out in Europe after all

Meta paused its launch in Europe last year.

www.theverge.com

Meta AI is rolling out in Europe after all​

A more limited version of Meta AI will head to WhatsApp, Facebook, Instagram, and Messenger this week.

by Emma Roth

Mar 20, 2025, 2:00 AM EDT

5 Comments

Illustration by Nick Barclay / The Verge

Emma Roth is a news writer who covers the streaming wars, consumer tech, crypto, social media, and much more. Previously, she was a writer and editor at MUO.

Meta is bringing its AI chatbot to Europe almost one year after pausing its launch in the region. Starting this week, Meta AI will roll out across WhatsApp, Facebook, Instagram, and Messenger in 41 European countries and 21 overseas territories — but it will be limited to text-based chat features for now.

Meta AI launched in the US in 2023. While the company intended to bring the assistant to Europe, Meta had to pause the rollout in the region after Ireland’s privacy watchdog asked it to delay training on content posted by Facebook and Instagram users. It also halted the launch of its multimodal Llama AI model in the European Union due to regulatory concerns.

For now, Meta says its AI assistant will only function as a chatbot for users in Europe, helping to brainstorm ideas, plan a trip, or answer specific questions using information from the web. European users will also be able to use Meta AI to surface certain kinds of content on their Instagram feed. However, they can’t use the tool to do things like generate or edit images, as well as ask questions about a photo. The model isn’t trained on EU user data, either.

“This launch follows almost a year of intensive engagement with various European regulators and for now, we are only offering a text-only model in the region which wasn’t trained on first-party data from users in the EU,” Meta spokesperson Ellie Heatrick tells The Verge. “We will continue to work collaboratively with regulators so that people in Europe have access to and are properly served by Meta’s AI innovations that are already available to the rest of the world.”

Last November, Meta started bringing some of its AI features to its Ray-Bans smart glasses in the EU, but the glasses currently don’t support multimodal features that let users ask Meta AI about what they can see. It doesn’t seem like Meta is giving up on plans to bring more features to the European version of Meta AI, as the company says it will work to “find parity with the US and expand our offering over time.”

 

[bnew]

The womb of oblivion.

The Body Scout author Lincoln Michel has a thoughtful literary analysis of Sam Altman’s AI-written “metafictional literary short story” — which, like a lot of AI text, scans well without exactly adding up: > “I haven’t actually seen anyone praise the story as a story. No one is lauding the...

www.theverge.com

The womb of oblivion. | The Verge​

Posted Mar 20, 2025 at 10:32 AM EDT

The womb of oblivion.

The Body Scout author Lincoln Michel has a thoughtful literary analysis of Sam Altman’s AI-written “metafictional literary short story” — which, like a lot of AI text, scans well without exactly adding up:

“I haven’t actually seen anyone praise the story as a story. No one is lauding the memorable characters or marveling at the vivid setting. Instead, the praise has focused on “good lines.” The purple prose. And the prose is the worst part.”

MFA vs. LLM: Is OpenAI's Metafiction Short Story Actually "Good"?
[countercraft.substack.com]