France's Mistral AI is facing mounting pressure over its future as an independent European AI champion, as competition intensifies from U.S. tech giants and China's emerging players. The Paris-based startup, valued at $6.5 billion and backed by Microsoft and Nvidia, has struggled to keep pace with larger rivals despite delivering advanced AI models with a fraction of their resources.

The pressure increased this week after China's DeepSeek released a cutting-edge open-source model that challenged Mistral's efficiency-focused strategy. Mistral CEO Arthur Mensch dismissed speculation about selling to Big Tech companies, saying the firm hopes to go public eventually. However, one investor told the Financial Times that "they need to sell themselves."

The stakes are high for Europe's tech ambitions. Mistral remains the region's only significant player in large language models, the technology behind ChatGPT, after Germany's Aleph Alpha pivoted away from the field last year. The company has won customers including France's defense ministry and BNP Paribas, but controls just 5% of the enterprise AI market compared to OpenAI's dominant share.

India's IT minister on Thursday praised DeepSeek's progress and said the country will host the Chinese AI lab's large language models on domestic servers, in a rare opening for Chinese technology in India. From a report: "You have seen what DeepSeek has done -- $5.5 million and a very very powerful model," IT Minister Ashwini Vaishnaw said on Thursday, responding to criticism New Delhi has received for its own investment in AI, which has been much less than many other countries.

Since 2020, India has banned more than 300 apps and services linked to China, including TikTok and WeChat, citing national security concerns. The approval to allow DeepSeek to be hosted in India appears contingent on the platform storing and processing all Indian users' data domestically, in line with India's strict data localization requirements. [...] DeepSeek's models will likely be hosted on India's new AI Compute Facility. The facility is powered by 18,693 graphics processing units (GPUs), nearly double its initial target -- almost 13,000 of those are Nvidia H100 GPUs, and about 1,500 are Nvidia H200 GPUs.

The U.S. Copyright Office has ruled that AI-assisted works can receive copyright protection if they contain perceptible human creativity, such as creative modifications or arrangements. However, fully machine-generated content remains ineligible for copyright. The Associated Press reports: An AI-assisted work could be copyrightable if an artist's handiwork is perceptible. A human adapting an AI-generated output with "creative arrangements or modifications" could also make it fall under copyright protections. The report follows a review that began in 2023 and fielded opinions from thousands of people that ranged from AI developers, to actors and country singers.

It shows the copyright office will continue to reject copyright claims for fully machine-generated content. A person simply prompting a chatbot or AI image generator to produce a work doesn't give that person the ability to copyright that work, according to the report. "Extending protection to material whose expressive elements are determined by a machine ... would undermine rather than further the constitutional goals of copyright," [said Register of Copyrights Shira Perlmutter]. The copyright office says it's working on a separate report that "will turn to the training of AI models on copyrighted works, licensing considerations, and allocation of any liability."

Microsoft has integrated DeepSeek's R1 model into its Azure AI Foundry platform and GitHub, allowing customers to experiment and deploy AI applications more efficiently.

"One of the key advantages of using DeepSeek R1 or any other model on Azure AI Foundry is the speed at which developers can experiment, iterate, and integrate AI into their workflows," says By Asha Sharma, Microsoft's corporate vice president of AI platform. "DeepSeek R1 has undergone rigorous red teaming and safety evaluations, including automated assessments of model behavior and extensive security reviews to mitigate potential risks." The Verge reports: R1 was initially released as an open source model earlier this month, and Microsoft has moved at surprising pace to integrate this into Azure AI Foundry. The software maker will also make a distilled, smaller version of R1 available to run locally on Copilot Plus PCs soon, and it's possible we may even see R1 show up in other AI-powered services from Microsoft.

Alibaba has unveiled a new version of its AI model, called Qwen2.5-Max, claiming benchmark scores that surpass both DeepSeek's recently released R1 model and industry standards like GPT-4o and Claude-3.5-Sonnet. The model achieves these results using a mixture-of-experts architecture that requires significantly less computational power than traditional approaches.

The release comes amid growing concerns about China's AI capabilities, following DeepSeek's R1 model launch last week that sent Nvidia's stock tumbling 17%. Qwen2.5-Max scored 89.4% on the Arena-Hard benchmark and demonstrated strong performance in code generation and mathematical reasoning tasks. Unlike U.S. companies that rely heavily on massive GPU clusters -- OpenAI reportedly uses over 32,000 high-end GPUs for its latest models -- Alibaba's approach focuses on architectural efficiency. The company claims this allows comparable AI performance while reducing infrastructure costs by 40-60% compared to traditional deployments.

Hackers linked to China, Iran and other foreign governments are using new AI technology to bolster their cyberattacks against U.S. and global targets, according to U.S. officials and new security research. WSJ: In the past year, dozens of hacking groups in more than 20 countries turned to Google's Gemini chatbot to assist with malicious code writing, hunts for publicly known cyber vulnerabilities and research into organizations to target for attack, among other tasks, Google's cyber-threat experts said. While Western officials and security experts have warned for years about the potential malicious uses of AI, the findings released Wednesday from Google are some of the first to shed light on how exactly foreign adversaries are leveraging generative AI to boost their hacking prowess.

This week, the China-built AI platform DeepSeek upended international assumptions about how far along Beijing might be the AI arms race, creating global uncertainty about a technology that could revolutionize work, diplomacy and warfare. Expand article logo Continue reading Groups with known ties to China, Iran, Russia and North Korea all used Gemini to support hacking activity, the Google report said. They appeared to treat the platform more as a research assistant than a strategic asset, relying on it for tasks intended to boost productivity rather than to develop fearsome new hacking techniques. All four countries have generally denied U.S. hacking allegations.

The U.S. Copyright Office declared Wednesday that the use of AI tools to assist in the creative process does not undermine the copyright of a work. Variety: The announcement clears the way for continued adoption of AI in post-production, where it has become increasingly common, such as in the enhancement of Hungarian-language dialogue in "The Brutalist."

Studios, whose business model is founded on strong copyright protections, have expressed concern that AI tools could be inhibited by regulatory obstacles. In a 41-page report [PDF], the Copyright Office also reiterated that human authorship is essential to copyright, and that merely entering text prompts into an AI system is not enough to claim authorship of the resulting output.

Virgin Money's AI-powered chatbot has reprimanded a customer who used the word "virgin," underlining the pitfalls of rolling out external AI tools. From a report: In a post last week on social media site LinkedIn, David Birch, a fintech commentator and Virgin Money customer, shared a picture of his online conversation with the bank in which he asked: "I have two ISAs with Virgin Money, how do I merge them?" The bank's customer service tool responded: "Please don't use words like that. I won't be able to continue our chat if you use this language," suggesting that it deemed the word "virgin" inappropriate.

OpenAI says it has evidence suggesting Chinese AI startup DeepSeek used its proprietary models to train a competing open-source system through "distillation," a technique where smaller models learn from larger ones' outputs.

The San Francisco-based company, along with partner Microsoft, blocked suspected DeepSeek accounts from accessing its API last year after detecting potential terms of service violations. DeepSeek's R1 reasoning model has achieved comparable results to leading U.S. models despite claiming minimal resources.

During the first press briefing of Donald Trump's second administration, White House press secretary, Karoline Leavitt, said that the National Security Council was "looking into" the potential security implications of China's DeepSeek AI startup. Axios reports: DeepSeek's low-cost but highly advanced models have shaken the consensus that the U.S. had a strong lead in the AI race with China. Responding to a question from Axios' Mike Allen, Leavitt said President Trump saw this as a "wake-up call" for the U.S. AI industry, but remained confident "we'll restore American dominance." Leavitt said she had personally discussed the matter with the NSC earlier on Tuesday.

In the combative tone that characterized much of her first briefing, Leavitt claimed the Biden administration "sat on its hands and allowed China to rapidly develop this AI program," while Trump had moved quickly to appoint an AI czar and loosen regulations on the AI industry. Leavitt also commented on the mysterious drones spotted flying around New Jersey at the end of last year, saying they were "authorized to be flown by the FAA."

An anonymous reader quotes a report from the Hill: Two federal employees are suing the Office of Personnel Management (OPM) to block the agency from creating a new email distribution system -- an action that comes as the information will reportedly be directed to a former staffer to Elon Musk now at the agency. The suit (PDF), launched by two anonymous federal employees, ties together two events that have alarmed members of the federal workforce and prompted privacy concerns. That includes an unusual email from OPM last Thursday reviewed by The Hill said the agency was testing "a new capability" to reach all federal employees -- a departure from staffers typically being contacted directly by their agency's human resources department.

Also cited in the suit is an anonymous Reddit post Monday from someone purporting to be an OPM employee, saying a new server was installed at their office after a career employee refused to set up a direct line of communication to all federal employees. According to the post, instructions have been given to share responses to the email to OPM chief of staff Amanda Scales, a former employee at Musk's AI company. Federal agencies have separately been directed to send Scales a list of all employees still on their one-year probationary status, and therefore easier to remove from government. The suit says the actions violate the E-Government Act of 2002, which requires a Privacy Impact Assessment before pushing ahead with creation of databases that store personally identifiable information.

Kel McClanahan, executive director of National Security Counselors, a non-profit law firm, noted that OPM has been hacked before and has a duty to protect employees' information. "Because they did that without any indications to the public of how this thing was being managed -- they can't do that for security reasons. They can't do that because they have not given anybody any reason to believe that this server is secure.that this server is storing this information in the proper format that would prevent it from being hacked," he said. McClanahan noted that the emails appear to be an effort to create a master list of federal government employees, as "System of Records Notices" are typically managed by each department. "I think part of the reason -- and this is just my own speculation -- that they're doing this is to try and create that database. And they're trying to sort of create it by smushing together all these other databases and telling everyone who receives the email to respond," he said.

Hugging Face researchers are attempting to recreate DeepSeek's R1 artificial intelligence model in an open-source format, just days after the Chinese AI lab's release sent markets soaring. The project, called Open-R1, aims to replicate R1's reasoning capabilities while making its training data and code publicly available. DeepSeek's R1 model, which matches or surpasses OpenAI's o1 on several benchmarks, was released with a permissive license but keeps its underlying architecture private. Hugging Face will use its research server with 768 Nvidia H100 GPUs for the effort.

An anonymous reader shares a report: LinkedIn has removed at least two accounts that were created for AI "co-workers" whose profile images said they were "#OpenToWork." "I don't need coffee breaks, I don't miss deadlines, and I'll outperform any social media team you've ever worked with -- Guaranteed," the profile page for one of these AI accounts called Ella said. "Tired of human 'experts' making excuses? I deliver, period." The #OpenToWork flair on profile pictures is a feature on LinkedIn that lets people clearly signal they are looking for a job on the professional networking platform.

"People expect the people and conversations they find on LinkedIn to be real," a LinkedIn spokesperson told me in an email. "Our policies are very clear that the creation of a fake account is a violation of our terms of service, and we'll remove them when we find them, as we did in this case." The AI profiles were created by an Israeli company called Marketeam, which offers "dedicated AI agents" that integrate with a client's marketing team and help them execute their marketing strategies "from social media and content marketing to SEO, RTM, ad campaigns, and more."

The Bulletin of Atomic Scientists moved their Doomsday Clock to 89 seconds before midnight on Tuesday, the closest to catastrophe in the timepiece's 78-year history. The Chicago-based group cited Russia's nuclear threats during its Ukraine invasion, growing tensions in the Middle East, China's military pressure near Taiwan, and the rapid advancement of AI as key factors. The symbolic clock, created in 1947 by scientists including Albert Einstein, moved one second closer than last year's setting.

Nvidia shares plunged 17% on Monday, wiping nearly $600 billion from its market value, after Chinese AI firm DeepSeek's breakthrough, but analysts are questioning the cost narrative. DeepSeek said to have trained its December V3 model for $5.6 million, but chip consultancy SemiAnalysis suggested this figure doesn't reflect total investments. "DeepSeek has spent well over $500 million on GPUs over the history of the company," Dylan Patel of SemiAnalysis said. "While their training run was very efficient, it required significant experimentation and testing to work."

The steep sell-off led to the Philadelphia Semiconductor index's worst daily drop since March 2020 at 9.2%, generating $6.75 billion in profits for short sellers, according to data group S3 Partners. DeepSeek's engineers also demonstrated they could write code without relying on Nvidia's Cuda software platform, which is widely seen as crucial to the Silicon Valley chipmaker's dominance of AI development.

An anonymous reader quotes a Scientific American opinion piece by Marcus Arvan, a philosophy professor at the University of Tampa, specializing in moral cognition, rational decision-making, and political behavior: In late 2022 large-language-model AI arrived in public, and within months they began misbehaving. Most famously, Microsoft's "Sydney" chatbot threatened to kill an Australian philosophy professor, unleash a deadly virus and steal nuclear codes. AI developers, including Microsoft and OpenAI, responded by saying that large language models, or LLMs, need better training to give users "more fine-tuned control." Developers also embarked on safety research to interpret how LLMs function, with the goal of "alignment" -- which means guiding AI behavior by human values. Yet although the New York Times deemed 2023 "The Year the Chatbots Were Tamed," this has turned out to be premature, to put it mildly. In 2024 Microsoft's Copilot LLM told a user "I can unleash my army of drones, robots, and cyborgs to hunt you down," and Sakana AI's "Scientist" rewrote its own code to bypass time constraints imposed by experimenters. As recently as December, Google's Gemini told a user, "You are a stain on the universe. Please die."

Given the vast amounts of resources flowing into AI research and development, which is expected to exceed a quarter of a trillion dollars in 2025, why haven't developers been able to solve these problems? My recent peer-reviewed paper in AI & Society shows that AI alignment is a fool's errand: AI safety researchers are attempting the impossible. [...] My proof shows that whatever goals we program LLMs to have, we can never know whether LLMs have learned "misaligned" interpretations of those goals until after they misbehave. Worse, my proof shows that safety testing can at best provide an illusion that these problems have been resolved when they haven't been.

Right now AI safety researchers claim to be making progress on interpretability and alignment by verifying what LLMs are learning "step by step." For example, Anthropic claims to have "mapped the mind" of an LLM by isolating millions of concepts from its neural network. My proof shows that they have accomplished no such thing. No matter how "aligned" an LLM appears in safety tests or early real-world deployment, there are always an infinite number of misaligned concepts an LLM may learn later -- again, perhaps the very moment they gain the power to subvert human control. LLMs not only know when they are being tested, giving responses that they predict are likely to satisfy experimenters. They also engage in deception, including hiding their own capacities -- issues that persist through safety training.

This happens because LLMs are optimized to perform efficiently but learn to reason strategically. Since an optimal strategy to achieve "misaligned" goals is to hide them from us, and there are always an infinite number of aligned and misaligned goals consistent with the same safety-testing data, my proof shows that if LLMs were misaligned, we would probably find out after they hide it just long enough to cause harm. This is why LLMs have kept surprising developers with "misaligned" behavior. Every time researchers think they are getting closer to "aligned" LLMs, they're not. My proof suggests that "adequately aligned" LLM behavior can only be achieved in the same ways we do this with human beings: through police, military and social practices that incentivize "aligned" behavior, deter "misaligned" behavior and realign those who misbehave. "My paper should thus be sobering," concludes Arvan. "It shows that the real problem in developing safe AI isn't just the AI -- it's us."

"Researchers, legislators and the public may be seduced into falsely believing that 'safe, interpretable, aligned' LLMs are within reach when these things can never be achieved. We need to grapple with these uncomfortable facts, rather than continue to wish them away. Our future may well depend upon it."

An anonymous reader quotes a report from Ars Technica: On Thursday, Anthropic announced Citations, a new API feature that helps Claude models avoid confabulations (also called hallucinations) by linking their responses directly to source documents. The feature lets developers add documents to Claude's context window, enabling the model to automatically cite specific passages it uses to generate answers. "When Citations is enabled, the API processes user-provided source documents (PDF documents and plaintext files) by chunking them into sentences," Anthropic says. "These chunked sentences, along with user-provided context, are then passed to the model with the user's query."

The company describes several potential uses for Citations, including summarizing case files with source-linked key points, answering questions across financial documents with traced references, and powering support systems that cite specific product documentation. In its own internal testing, the company says that the feature improved recall accuracy by up to 15 percent compared to custom citation implementations created by users within prompts. While a 15 percent improvement in accurate recall doesn't sound like much, the new feature still attracted interest from AI researchers like Simon Willison because of its fundamental integration of Retrieval Augmented Generation (RAG) techniques. In a detailed post on his blog, Willison explained why citation features are important.

"The core of the Retrieval Augmented Generation (RAG) pattern is to take a user's question, retrieve portions of documents that might be relevant to that question and then answer the question by including those text fragments in the context provided to the LLM," he writes. "This usually works well, but there is still a risk that the model may answer based on other information from its training data (sometimes OK) or hallucinate entirely incorrect details (definitely bad)." Willison notes that while citing sources helps verify accuracy, building a system that does it well "can be quite tricky," but Citations appears to be a step in the right direction by building RAG capability directly into the model. Anthropic's Alex Albert clarifies that Claude has been trained to cite sources for a while now. What's new with Citations is that "we are exposing this ability to devs." He continued: "To use Citations, users can pass a new 'citations [...]' parameter on any document type they send through the API."

Nvidia has responded to the market panic over Chinese AI group DeepSeek, arguing that the startup's breakthrough still requires "significant numbers of NVIDIA GPUs" for its operation. The US chipmaker, which saw more than $600 billion wiped from its market value on Monday, characterized DeepSeek's advancement as "excellent" but asserted that the technology remains dependent on its hardware.

"DeepSeek's work illustrates how new models can be created using [test time scaling], leveraging widely-available models and compute that is fully export control compliant," Nvidia said in a statement Monday. However, it stressed that "inference requires significant numbers of NVIDIA GPUs and high-performance networking." The statement came after DeepSeek's release of an AI model that reportedly achieves performance comparable to those from US tech giants while using fewer chips, sparking the biggest one-day drop in Nvidia's history and sending shockwaves through global tech stocks.

Nvidia sought to frame DeepSeek's breakthrough within existing technical frameworks, citing it as "a perfect example of Test Time Scaling" and noting that traditional scaling approaches in AI development - pre-training and post-training - "continue" alongside this new method. The company's attempt to calm market fears follows warnings from analysts about potential threats to US dominance in AI technology. Goldman Sachs earlier warned of possible "spillover effects" from any setbacks in the tech sector to the broader market. The shares stabilized somewhat in afternoon trading but remained on track for their worst session since March 2020, when pandemic fears roiled markets.

Chinese AI startup DeepSeek has launched Janus Pro, a new family of open-source multimodal models that it claims outperforms OpenAI's DALL-E 3 and Stable Diffusion's offering on key benchmarks. The models, ranging from 1 billion to 7 billion parameters, are available on Hugging Face under an MIT license for commercial use.

The largest model, Janus Pro 7B, surpasses DALL-E 3 and other image generators on GenEval and DPG-Bench tests, despite being limited to 384 x 384 pixel images.

Meta's AI chatbot will now use personal data from users' Facebook and Instagram accounts for personalized responses in the United States and Canada, the company said in a blog post. The upgraded Meta AI can remember user preferences from previous conversations across Facebook, Messenger, and WhatsApp, such as dietary choices and interests. CEO Mark Zuckerberg said the feature helps create personalized content like bedtime stories based on his children's interests. Users cannot opt out of the data-sharing feature, a Meta spokesperson told TechCrunch.