Latest news with #DeepSeek-R1
The Star
14-05-2025
- Health
- The Star
DeepSeek's AI in hospitals is ‘too fast, too soon', Chinese medical researchers warn
A paper warning of 'substantial clinical risk' from an overreliance on AI models makes a team of medical researchers a rare voice of caution. — SCMP A team of researchers in China has questioned hospitals' rapid adoption of DeepSeek, warning that it creates clinical safety and privacy risks, raising red flags over the rush to use the artificial intelligence (AI) start-up's cost-efficient open-source models. As of early March, at least 300 hospitals in China have started using DeepSeek's large language models (LLMs) in clinical diagnostics and medical decision support. The researchers warned that DeepSeek's tendency to generate 'plausible but factually incorrect outputs' could lead to 'substantial clinical risk', despite strong reasoning capabilities, according to a paper published last month in the medical journal JAMA. The team includes Wong Tien Yin, founding head of Tsinghua Medicine, a group of medical research schools at Tsinghua University in Beijing. The paper was a rare voice of caution in China against the overzealous use of DeepSeek. The start-up has become the nation's poster child for AI after its low-cost, high-performance V3 and R1 models captured global attention this year. DeepSeek did not immediately respond to a request for comment. According to Wong, an ophthalmology professor and former medical director at the Singapore National Eye Centre, and his co-authors, healthcare professionals could become overreliant on or uncritical of DeepSeek's output. This could result in diagnostic errors or treatment biases, while more cautious clinicians could face the burden of verifying AI output in time-sensitive clinical settings, they said. While hospitals often choose private, on-site deployment of DeepSeek models instead of cloud-based solutions to mitigate security and privacy risks, this approach presents challenges. It 'shifts security responsibilities to individual healthcare facilities', many of which lack comprehensive cybersecurity infrastructure, according to the researchers. In China, the combination of disparities in primary care infrastructure and high smartphone penetration also created a 'perfect storm' for clinical safety concerns, they added. 'Underserved populations with complex medical needs now have unprecedented access to AI-driven health recommendations, but often lack the clinical oversight needed for safe implementation,' the researchers wrote. The paper reflects the healthcare community's increasing scrutiny of LLM use in clinical and medical settings, as organisations across China accelerate adoption. Researchers from the Chinese University of Hong Kong also published a paper last month on the cybersecurity of AI agents, finding that most powered by mainstream LLMs were susceptible to attacks, with DeepSeek-R1 being the most vulnerable. The country has sped up the use of LLMs in the healthcare sector amid a boom in generative AI technologies. Last month, Chinese fintech giant Ant Group launched nearly 100 AI medical agents on its Alipay payments app. The agents are based on medical experts from China's top hospitals. Tairex, a start-up incubated at Tsinghua University, also began internal tests of a virtual hospital platform in November. The platform features 42 AI doctors covering 21 departments, including emergency, respiratory, paediatrics and cardiology. The company aimed to make the platform available to the general public this year, it said at the time. – South China Morning Post
Business Mayor
13-05-2025
- Science
- Business Mayor
ecoLogicStudio unveils AI system for environmentally responsive urbanism at venice biennale
At the 2025 Venice Architecture Biennale, ecoLogicStudio presents FundamentAI, a collaborative installation developed with the Synthetic Landscape Lab at Innsbruck University and the Urban Morphogenesis Lab at the Bartlett, UCL. The project explores the intersection of architecture, biotechnology, and artificial intelligence through the lens of Venice's unique lagoon ecosystem. Installed at the Arsenale, FundamentAI proposes a new model for urban design that includes real-time ecological data as a driver for architectural formation. The installation integrates environmental signals, particularly from lagoon microorganisms, with generative design processes enabled by multimodal AI systems. Its design references Venice's traditional wooden 'bricole' foundation poles, reinterpreted as bio-fabricated, biodegradable 3D printed columns embedded with responsive technology. The project employs a participatory design interface, allowing visitors to contribute images and texts that are interpreted by AI models to generate urban forms. This system incorporates several AI technologies: DeepSeek-R1 and GPT-4o for image and text analysis; FLUX.1-dev on ComfyUI for image generation; TRELLIS for 3D modeling; and Kling AI for animation output. These tools collectively produce adaptive, AI-mediated architectural responses that are informed by both user input and environmental data, such as acidity levels and microbial activity in the lagoon. FundamentAI installation at the 2025 Venice Architecture Biennale | all images courtesy of ecoLogicStudio and the Synthetic Landscape Lab Microbial Signals and AI Shape Immersive 3D printed installation Architecturally, the installation includes a full-scale immersive environment featuring 3D printed columns made from biodegradable material. These elements are designed using origination points from microbial and climatic signals in the Venetian lagoon. They respond to changing environmental conditions with subtle lighting effects powered by an AI video system, visually encoding data streams such as light, acidity, and microbial growth into atmospheric shifts within the space. Read More Gaia recycled upholstery fabric collection by Skopos FundamentAI, a collaborative initiative by architecture innovation firm ecoLogicStudio, Synthetic Landscape Lab at Innsbruck University, and the Urban Morphogenesis Lab at the Bartlett, UCL, aims to reframe foundational urban elements as symbiotic and ecologically attuned rather than static or anthropocentric. The installation treats the lagoon not only as a geographic context but also as an active participant in the design process. Ecological data is captured, processed, and translated into form, giving the lagoon agency within the architectural narrative. FundamentAI also introduces the 'Capsule Urbanism' concept, an approach to urban design that incorporates compact, modular, and ecologically responsive elements. The installation uses capsule-scale fabrication as a testbed for ideas that could scale to broader urban contexts, particularly in coastal areas facing environmental precarity. the installation explores architecture shaped by AI and microbial signals from the Venetian lagoon Expanding Public Participation in Design Through AI-Driven Tools One of the central ambitions of the project is to demonstrate how AI tools can expand public participation in urban planning. By scanning a QR code, visitors are invited to upload photographs of their surroundings in Venice. The system processes these visual inputs to generate speculative urban models that are visually and structurally informed by both user data and real-time environmental feedback from the lagoon. While rooted in Venice's history and material culture, FundamentAI is conceived as a globally relevant framework, particularly for cities in the Global South facing rapid urbanization and ecological vulnerability. The adaptive design method allows for responsive, data-integrated planning that accommodates non-human actors and environmental variability, potentially offering new strategies for climate-resilient development. Beyond its exhibition at the Venice Biennale, the project forms part of a broader research trajectory by ecoLogicStudio and its partners. Related upcoming projects include DeepForestCube, to be presented at the Triennale International Exhibition 2025, and with the BioLab, scheduled for launch at the Bundeskunsthalle in Bonn in June 2025. 3D printed biodegradable columns reinterpret Venice's traditional bricole foundations generative AI processes respond to real-time environmental data like acidity and light interaction between the bio-fabricated columns and the video projection close-up view of 3D printed sculptures in biopolymers subtle lighting shifts reflect microbial activity and lagoon chemistry in real time
South China Morning Post
02-05-2025
- Business
- South China Morning Post
Nvidia raises concerns about Huawei's growing AI chip capabilities with US lawmakers
Advertisement The issues were raised during a closed-door meeting between Nvidia executives and the US House of Representatives Foreign Affairs Committee on Thursday. Among the topics discussed were Huawei's AI chips and how restrictions on Nvidia's chips in China could make Huawei's chips more competitive. 'If DeepSeek-R1 had been trained on [Huawei chips] or a future open-source Chinese model had been trained to be highly optimised to Huawei chips, that would risk creating a global market demand for Huawei chips,' the senior staff source said. In a statement, Nvidia spokesman John Rizzo said 'Jensen met with the House Foreign Affairs Committee to discuss the strategic importance of AI as national infrastructure and the need to invest in US manufacturing. He reaffirmed Nvidia's full support for the government's efforts to promote American technology and interests around the world.' Nvidia's chips, which are central to developing chatbots, image generators and other AI systems, have been the target of US export controls since the first administration of US President Donald Trump. Nvidia has responded by designing chips for the Chinese market that have complied with the changing rules. Advertisement
Yahoo
01-05-2025
- Business
- Yahoo
Why Is Microsoft (MSFT) Stock Soaring Today
Shares of tech giant Microsoft (NASDAQ:MSFT) jumped 10.4% in the morning session after the company reported strong first-quarter 2025 results, with revenue and operating income both beating Wall Street estimates, driven by surging demand for cloud and AI services, signaling resilient enterprise spend amid broader tech budget scrutiny. Sales rose 13%, supported by broad-based strength across all business segments. While Productivity and Business Processes grew 10% and More Personal Computing rose 6%, the standout performance in Azure tipped the scales, thanks to increased customer adoption of AI workloads and infrastructure. The bottom line was equally strong. Operating income climbed 16%, outpacing revenue growth, with operating margins expanding across all three segments. This margin strength helped boost net income, pushing earnings past analysts' estimates. Overall, this was a solid quarter with key areas of upside. The shares closed the day at $425.19, up 8.4% from previous close. Is now the time to buy Microsoft? Access our full analysis report here, it's free. Microsoft's shares are quite volatile and have had 18 moves greater than 2.5% over the last year. But moves this big are rare even for Microsoft and indicate this news significantly impacted the market's perception of the business. The biggest move we wrote about over the last year was 3 months ago when the stock dropped 7.1% on the news that stocks heavily tied to the AI market took a hit after Chinese artificial intelligence startup DeepSeek released a new large language model (DeepSeek-R1) that ranks competitively on key global benchmarks (coding competitions, math evaluations), uses less advanced semiconductor chips, costs significantly less to build (at $5.5 million - excluding non-compute costs), and has already achieved strong adoption after topping the iPhone App Store for AI apps. Notably, the company also open-sourced this model, a move that might make it harder for rivals to justify huge upfront expenditures on hardware, software, and expertise to develop similar systems. Speaking at the World Economic Forum in Davos, Switzerland, Microsoft CEO Satya Nadella praised DeepSeek's efforts, calling the new model "super impressive" for its open-source design, efficient inference-time computing, and high compute efficiency. "We should take the developments out of China very, very seriously," he added. Nadella's comments suggest that upstarts like DeepSeek could reshape the competitive landscape of AI. DeepSeek's announcement disrupts long-held assumptions in key ways: 1) It undercuts the narrative that bigger budgets and access to top-tier chips are the only ways forward for AI development. 2) By using less advanced hardware, DeepSeek opens the door for innovators who face high chip costs or export restrictions, reaffirming they can still compete. 3.) The model's success questions the growth narrative of chipmakers like Nvidia, whose soaring valuations depend on the demand for cutting-edge, high-performance hardware. Overall, DeepSeek's model demonstrates that AI innovation is no longer a race fueled solely by how much you spend, but rather by how resourceful you can be with what you have. Microsoft is up 1.6% since the beginning of the year, and at $425.19 per share, it is trading close to its 52-week high of $467.56 from July 2024. Investors who bought $1,000 worth of Microsoft's shares 5 years ago would now be looking at an investment worth $2,436. Here at StockStory, we certainly understand the potential of thematic investing. Diverse winners from Microsoft (MSFT) to Alphabet (GOOG), Coca-Cola (KO) to Monster Beverage (MNST) could all have been identified as promising growth stories with a megatrend driving the growth. So, in that spirit, we've identified a relatively under-the-radar profitable growth stock benefiting from the rise of AI, available to you FREE via this link.
Business Insider
01-05-2025
- Business
- Business Insider
Cybersecurity execs face a new battlefront: 'It takes a good-guy AI to fight a bad-guy AI'
Generative artificial intelligence is a relatively new technology. Consequently, it presents new security challenges that can catch organizations off guard. Chatbots powered by large language models are vulnerable to various novel attacks. These include prompt injections, which use specially constructed prompts to change a model's behavior, and data exfiltration, which involves prompting a model thousands, maybe millions, of times to find sensitive or valuable information. These attacks exploit the unpredictable nature of LLMs, and they've already inflicted significant monetary pain. "The largest security breach I'm aware of, in monetary terms, happened recently, and it was an attack against OpenAI," said Chuck Herrin, the field chief information security officer of F5, a multicloud-application and security company. AI models are powerful but vulnerable Herrin was referencing DeepSeek, an LLM from the Chinese company by the same name. DeepSeek surprised the world with the January 20 release of DeepSeek-R1, a reasoning model that ranked only a hair behind OpenAI's best models on popular AI benchmarks. But DeepSeek users noticed some oddities in how the model performed. It often constructed its response similarly to OpenAI's ChatGPT and identified itself as a model trained by OpenAI. In the weeks that followed, OpenAI told the Financial Times it had evidence that DeepSeek had used a technique called "distillation" to train its own model by prompting ChatGPT. That evidence OpenAI said it had was not made public, and it's unclear whether the company will pursue the matter further. Still, the possibility caused serious concern. Herrin said DeepSeek was accused of distilling OpenAI's models down and stealing its intellectual property. "When the news of that hit the media, it took a trillion dollars off the S&P," he said. Alarmingly, it's well known that exploiting AI vulnerabilities is possible. LLMs are trained on large datasets and generally designed to respond to a wide variety of user prompts. A model doesn't typically "memorize" the data it's trained on, meaning it doesn't precisely reproduce the training data when asked (though memorization can occur; it's a key point New York Times' copyright infringement lawsuit against OpenAI). However, prompting a model thousands of times and analyzing the results can allow a third party to emulate a model's behavior, which is distillation. Techniques like this can also gain some insight into the model's training data. This is why you can't secure your AI without securing the application programming interface used to access the model and "the rest of the ecosystem," Herrin told Business Insider. So long as the API is available without appropriate safeguards, it can be exploited. To make matters worse, LLMs are a " black box." Training an LLM creates a neural network that gains a general understanding of the training data and the relationships between data in it. But the process doesn't describe which specific "neurons" in an LLM's network are responsible for a specific response to a prompt. That, in turn, means it's impossible to restrict access to specific data within an LLM in the same way an organization might protect a database. Sanjay Kalra, the head of product management at the cloud security company Zscaler, said: "Traditionally, when you place data, you place it in a database somewhere." At some point, an organization could delete that data if it wanted to, he told BI, "but with LLM chatbots, there's no easy way to roll back information." The solution to AI vulnerabilities is … more AI Cybersecurity companies are tackling this problem from many angles, but two stand out. The first is rooted in a more traditional, methodical approach to cybersecurity. "We already control authentication and authorization and have for a long time," Herrin said. He added that while authenticating users for an LLM "doesn't really change" compared with authenticating for other services, it remains crucial. Kalra also stressed the importance of good security fundamentals, such as access control and logging user access. "Maybe you want a copilot that's only available for engineering folks, but that shouldn't be available for marketing, or sales, or from a particular location," he said. But the other half of the solution is, ironically, more AI. LLMs' "black box" nature makes them tricky to secure, as it's not clear which prompts will bypass safeguards or exfiltrate data. But the models are quite good at analyzing text and other data, and cybersecurity companies are taking advantage of that to train AI watchdogs. These models position themselves as an additional layer between the LLM and the user. They examine user prompts and model responses for signs that a user is trying to extract information, bypass safeguards, or otherwise subvert the model. "It takes a good-guy AI to fight a bad-guy AI," Herrin said. "It's sort of this arms race. We're using an LLM that we purpose-built to detect these types of attacks." F5 provides services that allow clients to use this capability both when deploying their own AI model on premises and when accessing AI models in the cloud. But this approach has its difficulties, and cost is among them. Using a security-tuned variant of a large and capable model, like OpenAI's GPT-4.1, might seem like the best path toward maximum security. However, models like GPT-4.1 are expensive, which makes the idea impractical for most situations. "The insurance can't be more expensive than the car," Kalra said. "If I start using a large language model to protect other large language models, it's going to be cost-prohibitive. So in this case, we see what happens if you end up using small language models." Small language models have relatively few parameters. As a result, they require less computation to train and consume less computation and memory when deployed. Popular examples include Meta's Llama 3-8B and Mistral's Ministral 3B. Kalra said Zscaler also has an AI and machine learning team that trains its own internal models. As AI continues to evolve, organizations face an unexpected security scenario: The very technology that suffers vulnerabilities has become an essential part of the defense strategy against those weak spots. But a multilayered approach, which combines cybersecurity fundamentals with security-tuned AI models, can begin to fill the gaps in an LLM's defenses.
