Despite the ruling, Musk’s lawsuit against OpenAI will proceed, with both parties agreeing to an expedited trial scheduled for this fall.

Musk vs. OpenAI

Musk co-founded OpenAI in 2015 with Sam Altman and others, initially as a nonprofit organization aimed at developing AI for the benefit of humanity. OpenAI quickly gained recognition as one of the most innovative AI startups at the time. Musk then contributed tens of millions of dollars to the cause before stepping down from the board in 2018, citing his commitments as Tesla’s CEO.

As TechnologyAdvice staff writer Megan Crouse wrote on eWeek in February 2025, “OpenAI’s corporate structure is unusual. Two for-profit entities, OpenAI GP LLC and OpenAI Global LLC, are owned by OpenAI, Inc, the nonprofit dedicated to ‘ensuring that safe artificial intelligence is developed and benefits all of humanity.” In September 2024, OpenAI announced it would restructure so the nonprofit has a minority stake in the for-profit companies.”

OpenAI eventually secured billions in funding; it has also gained the reputation as one of the top AI companies globally. Musk claimed this shift breached the company’s founding principles, prompting him to file a lawsuit in February 2024. The legal challenge argues that OpenAI’s leadership violated their original commitments and misused funding intended for nonprofit AI research.

OpenAI’s defense

In February 2025, a group led by Musk reportedly offered $97.4 billion to purchase OpenAI outright — a bid that Altman and the company rejected.

According to OpenAI, its embrace of a for-profit model is needed in order to secure the necessary capital for future AI development. Their team argued that Musk’s lawsuit is just a ploy meant to slow the ongoing development of the OpenAI platform, which is a major competitor to Musk’s xAI brand and its large language models (LLMs) – including the Grok AI platform.

Preparing for the legal battle ahead

Both Musk and Altman face a contentious legal fight that could reshape AI governance and corporate ethics. Musk’s legal team welcomed the decision to fast-track the trial, but the denied injunction hands OpenAI an early victory.

The post OpenAI Scores Early Legal Win Against Musk Over For-Profit Shift appeared first on eWEEK.

What is this AI phone?

Deutsche Telekom’s upcoming AI Phone is still a bit of a mystery; the development team has yet to release technical specifications to the public. It appears that the initial concept renderings featured an Android operating system. Several generative AI tools, including Perplexity Assistant, will come pre-installed and fully integrated into the device. The AI Phone will reportedly carry a price tag of less than $1,000, though the exact price hasn’t been announced.

“The trusted AI companion, our AI Phone, will help you in many situations: find reliable answers with reference to the source. Conveniently book a restaurant or taxi. Let an AI assistant do your shopping. All without having to switch between apps. Intuitively and preferably by voice. This is the future of AI innovation for consumers,” said Claudia Nemat, member of Deutsche Telekom’s Board of Management for Technology and Innovation department, in the press release.

But wait, there’s more to come

Smartphone users who don’t want to purchase a new device can download Perplexity and other types of AI models from the MeinMagenta app later this year. The app is compatible with Android and iOS hardware.

Even more features, upgrades, and updates will be coming to the MeinMagenta app with Magenta AI – some have already been slated for release this summer.

• Google Cloud AI: Snap photos or record video on your phone and use Google Cloud AI to provide more context.
• ElevenLabs: Create AI-generated podcasts based on short prompts, URLs, or documents.
• Picsart: Use AI to transform your images with various styles and themes.

But the tools mentioned above are just the beginning. We’ll likely see more additions from Deutsche Telekom and its partners, including some of the top AI companies, in the near future.

Pushing the boundaries of AI technology

The developers and engineers driving AI innovation are constantly pushing the boundaries of what their technology can do. While most consumer usage of AI revolves around simple text, image, video, or audio generation, the new AI Phone takes it one step further with seamless, AI-assisted restaurant bookings, online shopping, real-time language translation, and much more.

The post Deutsche Telekom: New Perplexity AI Phone For Under $1K appeared first on eWEEK.

What is DeepSeek?

Based in Hangzhou, Zhejiang, China, DeepSeek was developed by a company of the same name. The AI platform was launched in November 2023, and since then, its developers have rolled out multiple generative AI tools.

• DeepSeek Coder: The first model made available to the public, DeepSeek Coder mimicked the functionality of Llama’s generative AI.
• DeepSeek-LLM: Released a few weeks later, DeepSeek-LLM is also based on Llama’s architecture.
• DeepSeek-MoE: This version implemented the MoE (mixture of experts) technique to supplement DeepSeek’s machine learning capabilities.
• DeepSeek-Math: As the name suggests, DeepSeek-Math uses AI to solve complex mathematical equationsI.
• DeepSeek-V2: The second version of DeepSeek was released in May 2024. It includes updated versions of Coder, Chat, Lite, and Lite-Chat.
• DeepSeek-V3: While an updated version of DeepSeek was released in December 2024, the architecture basically mimics that of V2.
• DeepSeek-R1: Unveiled in January 2025, R1 is the most powerful iteration of DeepSeek we’ve seen to date.

When DeepSeek claims its AI models could achieve 545% profit margins, the company is referring to this entire lineup. But some releases, like DeepSeek-R1, are expected to generate more profit than others.

Driving profits with DeepSeek-V3 and DeepSeek-R1

According to a recent post on X, DeepSeek’s team indicated that the cost of inferencing to sales for their V3 and R1 models resulted in total profit margins of 545%. In this context, inferencing describes the total cost of resources, such as data storage and electricity, involved in developing their large language models (LLMs).

However, the company clarified that these numbers do not reflect actual revenue. These figures don’t factor in activities related to training or research and development, either, and those costs have not been made public.

Maximizing ROI in the AI boom

Despite being less than two years old, DeepSeek has already positioned itself as a major AI contender. In a time when most AI developers are still trying to determine the financial feasibility of their AI platforms, DeepSeek is flirting with profit margins of 545%. Although they have yet to realize these profits, the fact that they’re already discussing numbers this high means that the AI company is definitely moving in the right direction.

Read about three possible cybersecurity risks with DeepSeek according to experts.

The post AI Models Drive DeepSeek’s 545% Profit Margin Projection appeared first on eWEEK.

Hunyuan Turbo S is capable of replying to user inputs and queries within one second, which is even faster than DeepSeek-R1, according to the company and as reported by Reuters. We haven’t found any speed benchmarks that confirm Tencent’s claim.

How Hunyuan Turbo S compares to competitors in the benchmarks

According to benchmarks provided by Tencent as reported by WinBuzzer, Hunyuan Turbo S leads many competitors in a variety of areas. The following benchmarks are commonly used to evaluate the functionality, efficiency, and accuracy of large language models (LLMs).

• Chinese: Hunyuan Turbo S ranks the highest in Chinese language benchmarks performed by CMMLU, but DeepSeek-R1-Zero leads in C-Eval’s benchmarks.
• Alignment: Although Hunyuan Turbo S outperforms GPT-4o, Claude 3.5, Llama 3.1, and DeepSeek-V3 in benchmarks from LiveBench, it lags slightly behind Claude 3.5 in benchmarks from IF-Eval.

Some of Hunyuan Turbo S’s weaknesses include:

• Math: Hunyuan Turbo S outperforms GPT-4o, Claude 3.5, Llama 3.1, and DeepSeek-V3 in some benchmarks, but DeepSeek-R1-Zero leads them all as scored by AIME 2024 and MATH.
• Knowledge: Hunyuan Turbo S ranks fairly high on most knowledge benchmarks, but it doesn’t quite match up to DeepSeek-R1-Zero in the benchmarks from MMLU, MMLU-Pro, and SimpleQA.
• Reasoning: Hunyuan Turbo S only ranks third-highest, behind GPT-4o and Claude 3.5, on BBH’s reasoning benchmarks.
• Code: While HumanEval has Hunyuan Turbo S sitting right behind Claude for coding capabilities, it’s a bit further behind DeepSeek-V3, DeepSeek-R1-Zero, and GPT-4o on LiveCodeBench’s results.

While Hunyuan Turbo S is the clear winner in certain cases, it still falls behind DeepSeek-R1-Zero in several instances.

A strong competitor in the AI race

Tencent’s new Hunyuan Turbo S platform solidifies the Chinese tech giant’s position in the race to develop the fastest and most powerful AI platform. Although it’s not Tencent’s first foray into the world of generative AI tools, it is the company’s most noteworthy entry to date – and it’s certainly one to watch over the coming weeks, months, and years.

The post Tencent’s New DeepSeek Competitor Looks Promising Based on Key AI Benchmarks  appeared first on eWEEK.

On the flip side, 31% say AI will not have any impact on the workplace. And, workers who are already using AI in finance, banking, accounting, insurance, real estate, and technology are more likely to believe they’ll see increased job opportunities as a result of AI.

Who is using AI in the workplace?

According to Pew’s research, 16% of U.S. workers use AI for at least part of their jobs, and 57% of current AI usage in the workplace involves research-oriented chatbots.

Additional findings include:

• 40% of AI chatbot users agree that AI chatbots and large language models (LLMs) help them perform tasks more quickly.
• Only 29% of AI chatbot users believe that AI chatbots have improved the quality of their work.
• Workers aged 18 to 29 make up the largest group of AI users, with only 17% older than 29.
• 73% of those who currently use AI in the workplace are under the age of 50.
• 51% of AI users in the workplace hold a bachelor’s degree or higher.

Despite these findings, 55% of workers say they rarely or never use AI chatbots in the workplace. It seems that the technology hasn’t been quite as disruptive as some might think.

Realizing the full potential of an AI-driven workplace

Pew’s latest survey underscores some of the biggest and most common concerns with AI. While it’s easy to conclude that U.S. workers are pessimistic about the future of AI in the workplace, it’s important to remember that most types of AI models are still in their infancy stages of development and, as such, we have yet to see their full potential.

The post 52% of US Workers Fear AI Will Impact Their Jobs, According to Pew Research appeared first on eWEEK.

While the video is clearly AI-generated, questions remain about the identity of the prankster, the generative AI tools used to create it, and how the perpetrator managed to hijack the department’s television systems.

Reuters reported that department spokesperson Kasey Lovett said in an email, “Another waste of taxpayer dollars and resources. Appropriate action will be taken for all involved.”

Coincidental timing of Trump-Musk prank and RTO mandate?

The incident might seem like a random act of civil disobedience; however, the timing suggests that this was a deliberate and well-executed prank carefully staged on a very specific date for maximum impact.

The hijacking occurred on Monday, Feb. 24, which was the same day a nationwide return-to-work mandate took effect for federal employees. With no immediate way to trace the source of the unauthorized broadcast, staff had no choice but to manually disconnect screens to stop the video from playing.

A Reuters/IPSOS poll released just a day before the incident found that only 42% of Americans surveyed support the newly created Department of Government Efficiency, or DOGE, and its recent initiatives.

Spotting AI-generated deepfakes

Observers quickly spotted a glaring flaw in the video: Musk appeared to have two left feet, a common telltale sign of AI-generated content. But that didn’t stop the video from spreading rapidly across major social media platforms and news outlets.

While this particular deepfake may have been intended as a prank, it highlights some serious concerns with AI and, in particular, deepfake content. Experts have yet to determine which AI tools were used to create the video, but the increasing accessibility of modern AI software means that nearly anyone can create their own deepfakes with the best AI video generators.

While the Trump-Musk video is relatively easy to debunk, more sophisticated deepfakes are becoming harder to detect. Most deepfakes have some telltale signs of AI usage — including distorted facial features, unnatural movements, or inconsistent hand and foot placement — but as the technology improves, distinguishing fakes from reality will likely become increasingly challenging.

The post Deepfake Video of Trump and Elon Musk Raises Concerns About Gen AI & Security appeared first on eWEEK.

Why is Microsoft canceling data center leases?

The report begins by highlighting the cancellation of “a couple hundred MWs” worth of data center capacity leasing agreements. Although the report doesn’t name any companies, it confirms that these cancellations involved at least two separate data center operators. Considering that Microsoft is one of the top AI companies, the cancelled leases could be devastating to these private data center operators and their bottom lines.

While the authors of the report aren’t certain of the cause for these terminations, they point to a potential overcommitment on behalf of Microsoft. They also suggest that Microsoft may be citing various technical delays to justify the canceled leases.

Uncertainty with Microsoft’s potential conversions

Microsoft is also shying away from converting some of its current Statements of Qualification, also known as SOQs, into leases; however, TD Cowen’s report does not clarify if these conversions are simply delayed or fully cancelled. Since the Statement of Qualification usually serves as a precursor to a full-on lease, companies that were expecting to sign leases in the future now face uncertainty.

Microsoft’s reallocation of resources

Microsoft seems to be reallocating resources from international data center investments and into the U.S. market. According to TD Cowen, the reallocated funds could indicate a slowdown with international leasing contracts.

Per the report, Microsoft was the most active lessee in terms of AI data center capacity throughout all of 2023 and into the first half of 2024. Much of this activity was earmarked specifically to support Azure’s OpenAI workloads and the usage of different types of AI models — including GPT-4o, DALL-E, and others. A slowdown could be a sign of Microsoft adjusting its data center strategy to better align with demand.

Interpretation of Microsoft’s actions

The authors of the TD Cowen report were quick to point out that their conclusions are an interpretation of Microsoft’s recent actions, and that their forecast is subject to change as more information becomes available.

Bloomberg reported that a Microsoft spokesperson said in the company’s statement: “While we may strategically pace or adjust our infrastructure in some areas, we will continue to grow strongly in all regions. Our plans to spend over $80 billion on infrastructure this FY remains on track as we continue to grow at a record pace to meet customer demand.”

Also, Microsoft CEO Satya Nadella has expressed his excitement for AI data center spending and growth throughout 2028.

The post Microsoft Cancels Some AI Data Center Leases Reportedly, Despite $80B Pledge appeared first on eWEEK.

Gemini AI at a glance

Unveiled in December 2023, Google’s Gemini, which ranks amongst the top generative AI tools on the market, is the company’s most capable LLM to date. The platform has seen numerous updates since its release, with the latest release, version 2.0, building on the foundation established by its predecessors. As such, Gemini 2.0 Flash, as it’s officially known, outperforms the benchmarks set with Gemini 1.5 Pro – and it does so at double the speed.

Gemini is a multimodal AI platform. Not only does it support text, image inputs, audio, and video inputs, it’s capable of generating and outputting content in these formats, too. The AI can also use Google Search, access third-party functions, and execute code. Supported coding languages include C++, Go, Java, JavaScript, and Python.

Embracing the AI agentic era

All of the technology packed into Gemini paves the way for the use of AI agents, which are microsized apps that utilize AI to help users complete various tasks and make decisions. Developers with Gemini are also leveraging spatial reasoning within next-gen robotics to bring Gemini’s capabilities into the physical world.

Some of the most popular agents built into the latest version of Gemini include:

• Project Astra: This AI-powered virtual assistant includes support for multiple spoken languages.
• Project Mariner: This research prototype integrates with your web browser to process and understand text, code, images, and even individual pixels.
• Jules: This AI-driven code agent assists developers with coding projects.
• Genie 2: This AI model can develop 3D games using nothing more than a single image as its prompt.

The future of Google’s Gemini

Although some users will likely be hesitant to make the switch, the Gemini platform is still early enough in its development cycle that it will likely continue to receive updates – and new users – for years to come.

The post Google Removes Gemini From Its Main App for iOS Devices appeared first on eWEEK.

“Laboratory science is resource-intensive,” said Professor José Penadés, a member of Imperial College London’s Department of Infectious Diseases and co-leader of the recent AI co-scientist project. “With global challenges like antimicrobial resistance looming, it’s clear we need to do more with less and speed up new discoveries.”

Developing the AI co-scientist

Google developed its AI co-scientist through a partnership with Imperial College London and the Fleming Initiative. The multi-agent AI platform was built on the Gemini 2.0 framework, which already cemented Google’s place among the top AI companies. But it’s important to note that Google isn’t using the system in an attempt to automate the traditional scientific process. Instead, the company is focused on building one of the most powerful AI collaboration tools to date.

While the AI co-scientist was developed over the course of the past few months, the Google team only recently released the initial test findings to the general public. Specifically, the teams from Google, Imperial, and the Fleming Initiative tested AI-generated hypotheses across three separate biomedical applications:

• Drug repurposing
• Novel treatment target discovery
• Understanding mechanisms of gene transfer evolution

The system can be used in a variety of ways. Not only can researchers use the AI co-scientist to elaborate on their own seed ideas, but they can also provide feedback on any of the platform’s generated outputs. In addition, Google’s AI co-scientist can perform online searches and access other types of AI models as necessary.

Joining the Trusted Tester Program

Researchers and organizations that want to use Google’s new AI co-scientist in its infancy stages can express their interest by signing up for Google’s Trusted Tester Program. Although the platform isn’t currently accessible, those who are accepted into the program will receive regular updates as new parts of the AI co-scientist are made available.

Accelerating Future Breakthroughs

The ultimate goal of Google’s new AI co-scientist is twofold: to enhance collaborative research and accelerate future scientific breakthroughs. Although there are still some limitations and imperfections to address, the multi-agent system has already shown promising results during its initial testing phase.

Learn more about the role of generative AI in healthcare.

The post Google’s AI Co-Scientist Speeds Discoveries by Helping Researchers “Do More With Less” appeared first on eWEEK.

HP’s acquisition includes Humane’s AI-driven platform Cosmos, as well as other intellectual properties and key staff members from the company who will form HP’s new AI Innovation lab called HP IQ.

Introducing HP’s new AI lab

HP IQ builds on the foundation established by the Humane team and Cosmos. The new AI lab will utilize a team of data engineers and architects from Humane in the development of an intelligent ecosystem meant to support future HP products and services.

“We’re excited to join HP at such a pivotal moment in the industry and help shape the future of intelligent experiences. HP’s scale, global reach, and operational excellence – combined with our design-led approach, integration technology, and engineering expertise – will redefine workforce productivity,” said Humane’s co-founders Bethany Bongiorno and Imran Chaudhri in a joint statement.

What is the Ai Pin, and what does this news mean for its users?

Named one of Time’s Best 200 Inventions of 2023, Humane’s Ai Pin is wearable tech that clips onto the user’s shirt. Not only does it serve as a cellular phone, digital camera, voice assistant, and a multilingual translator, it also includes a camera and a touchpad. Wearers of the Ai Pin can even use the device to project a small screen onto their own hand, which is a rather unique, albeit limited, feature.

According to the latest update from Humane, consumer availability of purchasing the Ai Pin is being discontinued immediately. Those who already own an Ai Pin will be able to access its full range of functionality until February 28, 2025, at which point Humane’s servers will be shuttered.

Humane will delete all of its remaining consumer data at that time. Users who want to save any data from their Ai Pins are encouraged to sync their devices and download their data by the end of the month.

The post HP’s $116M Acquisition Discontinues Humane’s Ai Pin in Favor of New AI Lab to “Redefine Workforce Productivity” appeared first on eWEEK.