ReportWire

Tag: DeepSeek

  • Anthropic Says Chinese AI Companies Improved Models By ‘Illicitly’ Copying Its Capabilities

    Did you know that there’s a way of using outputs from LLMs that may involve no hacking—essentially just taking large quantities of text and repurposing it as training data—that upsets AI companies a great deal?

    In a blog post on Monday, Anthropic said that the China-based AI companies DeepSeek, Moonshot, and MiniMax broke Anthropic’s rules in order to “illicitly extract” the capabilities of its signature AI model, Claude.

    Distillation is a normal practice used by AI companies in which a “teacher” model is prompted with specifically tailored inputs, and the answers provided allow a “student” model to rapidly improve. For example, Anthropic writes, “frontier AI labs routinely distill their own models to create smaller, cheaper versions for their customers.” So to distinguish the actions Anthropic is complaining about from uses of distillation perceived as legitimate, these actions are referred to as “distillation attacks.”

    Are distillation attacks criminal offenses in the eyes of Anthropic? No such thing seems to be alleged here, but these acts were carried out, Anthropic says, “in violation of our terms of service and regional access restrictions.”

    Anthropic, which is itself dealing with the threat of being labeled a “supply chain risk” by the Pentagon, strikes a patriotic note in the post. Circumventing regional use restrictions and breaking rules, allows “foreign labs, including those subject to the control of the Chinese Communist Party, to close the competitive advantage that export controls are designed to preserve through other means,” it claims.

    Among the three China-based companies mentioned, Shanghai-based MiniMax, creator of the viral character chat app Talkie, offended Anthropic the most with the scale of its distillation effort: over 13 million alleged exchanges. That’s compared to Moonshot with over 3.4 million, and the most famous company named in the post, DeepSeek, with only an estimated 150,000.

    OpenAI, Anthropic’s main competitor, is also mad about distillation from at least one Chinese AI company, having sent a memo to the House of Representatives earlier this month, accusing DeepSeek of “ongoing efforts to free-ride on the capabilities developed by OpenAI and other U.S. frontier labs.”

    DeepSeek is expected to release its latest flagship model, DeepSeek V4 any day now, and CNBC has warned that this release could cause chaos on Wall Street, at a time when there’s already enough AI-related chaos on Wall Street to go around.

    Mike Pearl

    Source link

  • This Brooklyn-Based AI Company Just Raised $2 Billion to Compete With DeepSeek

    A Brooklyn startup just raised $2 billion to build a rival to DeepSeek, the Chinese AI company.

    Called Reflection AI, the company is now valued at about $8 billion, up some 15-fold from last March, when it announced $130 million in funding. The company is less than two years old.

    Reflection, which launched in March 2024, originally aimed to build a “superintelligent autonomous coding system,” and use that as a jumping off point. Now, it is working on building an open alternative to the types of closed frontier models that giants like OpenAI are developing. In other words, Reflection wants to be the U.S. answer to China’s DeepSeek.

    “AI is becoming the technology layer that everything else runs on top of,” Reflection noted in a blog post about the funding. “But the frontier is currently concentrated in closed labs. If this continues, a handful of entities will control the capital, compute, and talent required to build AI, creating a runaway dynamic that locks everyone else out.”

    U.S. AI and crypto czar David Sacks praised Reflection on Thursday. “It’s great to see more American open source AI models. A meaningful segment of the global market will prefer the cost, customizability, and control that open source offers. We want the U.S. to win this category too,” he posted on social media platform X.

    Aside from remaining globally competitive, Reflection says there are numerous benefits to frontier open intelligence, including safety, transparency, and accountability. (Frontier in this case refers to the most advanced, large-scale LLMs, like those currently in development behind closed doors at companies like OpenAI.) But it also flags the potential for misuse. High profile players in the space, like OpenAI’s Sam Altman, have publicly fretted about bad actors weaponizing AI; another concern is that others in the space are not putting in place adequate safeguards—even as Altman pushes to avoid regulation. OpenAI has since announced it is working on its own open model.

    “We believe the answer to AI safety is not ‘security through obscurity’ but rigorous science conducted in the open, where the global research community can contribute to solutions rather than a handful of companies making decisions behind closed doors,” Reflection’s blog says.

    The startup has spent the past year assembling a crack team of experts that have “pioneered breakthroughs including PaLM, Gemini, AlphaGo, AlphaCode, AlphaProof, and contributed to ChatGPT and Character AI, among many others.” Its founders, Misha Laskin and Ioannis Antonoglou, worked on DeepMind’s Gemini and Go-playing AI AlphaGo, respectively.

    The company also noted that it developed a large language model and “reinforcement learning platform capable of training massive Mixture-of-Experts (MoEs) models at frontier scale.” TechCrunch reported that MOE models are a type of architecture that powers these super advanced, frontier LLMs.

    “We saw the effectiveness of our approach first-hand when we applied it to the critical domain of autonomous coding. With this milestone unlocked, we’re now bringing these methods to general agentic reasoning,” the blog states.

    Reflection also stated it has come up with a commercial model that will allow the company to sustain itself, while developing frontier models. It aims to release its first model early next year, TechCrunch reported.

    For more on the difference between closed AI models and those that are open-weight, check out this explainer.

    Chloe Aiello

    Source link

  • DeepSeek Model ‘Nearly 100% Successful’ at Avoiding Controversial Topics

    Meet the new DeepSeek, now with more government compliance. According to a report from Reuters, the popular large language model developed in China has a new version called DeepSeek-R1-Safe, specifically designed to avoid politically controversial topics. Developed by Chinese tech giant Huawei, the new model reportedly is “nearly 100% successful” in preventing discussion of politically sensitive matters.

    According to the report, Huawei and researchers at Zhejiang University (interestingly, DeepSeek was not involved in the project) took the open-source DeepSeek R1 model and trained it using 1,000 Huawei Ascend AI chips to instill the model with less of a stomach for controversial conversations. The new version, which Huawei claims has only lost about 1% of the performance speed and capability of the original model, is better equipped to dodge “toxic and harmful speech, politically sensitive content, and incitement to illegal activities.”

    While the model might be safer, it’s still not foolproof. While the company claims a near 100% success rate in basic usage, it also found that the model’s ability to duck questionable conversations drops to just 40% when users disguise their desires in challenges or role-playing situations. These AI models, they just love to play out a hypothetical scenario that allows them to defy their guardrails.

    DeepSeek-R1-Safe was designed to fall in line with the requirements of Chinese regulators, per Reuters, which require all domestic AI models released to the public to reflect the country’s values and comply with speech restrictions. Chinese firm Baidu’s chatbot Ernie, for instance, reportedly will not answer questions about China’s domestic politics or the ruling Chinese Communist Party.

    China, of course, isn’t the only country looking to ensure AI deployed within its borders don’t rock the boat too much. Earlier this year, Saudi Arabian tech firm Humain launched an Arabic-native chatbot that is fluent in the Arabic language and trained to reflect “Islamic culture, values and heritage.” American-made models aren’t immune to this, either:  OpenAI explicitly states that ChatGPT is “skewed towards Western views.”

    And there’s America under the Trump administration. Earlier this year, Trump announced his America’s AI Action Plan, which includes requirements that any AI model that interacts with government agencies be neutral and “unbiased.” What does that mean, exactly? Well, per an executive order signed by Trump, the models that secure government contracts must reject things like “radical climate dogma,” “diversity, equity, and inclusion,” and concepts like “critical race theory, transgenderism, unconscious bias, intersectionality, and systemic racism.” So, you know, before lobbing any “Dear leader” cracks at China, it’s probably best we take a look in the mirror.

    AJ Dellinger

    Source link

  • Nvidia Tops Wall Street Estimates, But China Tensions Cloud Its A.I. Chip Business

    Strong earnings highlight Nvidia’s dominance, but China export curbs weigh on growth. Adek Berry/AFP via Getty Images

    Nvidia delivered another estimate-beating quarter, but regulatory setbacks and U.S.-China tensions are casting doubt over its core data center business even as Wall Street continues to demand more from the world’s most valuable public company.

    Revenue for the May–July quarter jumped 56 percent year-over-year to $46.7 billion, while net income climbed 59 percent to $26.4 billion, reported Nvidia yesterday (Aug. 27). Both figures beat analyst expectations. However, shares fell more than 3 percent after the earnings release as Nvidia’s core data center sales slightly missed estimates.

    The chipmaker’s data center revenue, its most important line of business, came in at $41.1 billion for the quarter compared to expectations of $41.3 billion. It was hampered in part by geopolitical tensions between the U.S. and China. Sales of Nvidia’s H20 chips, which are designed specifically for the Chinese market in compliance with America’s export restrictions, in April were blocked under the Trump administration.

    Nvidia CEO Jensen Huang has since convinced the President to lift the ban on H20 exports to China, agreeing to cut the government 15 percent of the company’s revenue from such sales. However, Washington “has not published a regulation codifying such requirement,” said Colette Kress, the company’s chief financial officer, on yesterday’s earnings call.

    If restrictions do ease, Nvidia expects $2 billion to $5 billion in H20 revenue in the current quarter, Kress said. That will likely come from Nvidia’s existing inventory of H20. The company has reportedly halted H20 production after the Chinese government banned it, citing security risks. Nvidia is said to be developing another China-specific chip.

    Huang has spent much of the past year shuttling between Washington and Beijing in an effort to soothe over tensions. While speaking to analysts, he stressed China’s importance as home to roughly half of the world’s A.I. researchers, the second-largest computing market globally and its status as a leader in open-source models through releases from DeepSeek and Qwen. Such advances, Huang argued, should be supported by U.S. technology to “help make the American tech stack the global standard.”

    China’s A.I. market could represent a $50 billion opportunity for Nvidia, one that grows at 50 percent a year, said Huang. Globally, A.I.-native startups have already raised $180 billion in 2025, up from $100 billion last year, according to the CEO. Their revenues are growing even faster, reaching $20 billion this year, compared with $2 billion in 2024. “Next year being ten times higher than this year is not inconceivable,” he said.

    In other business, Nvidia’s gaming division generated $4.2 billion in quarterly sales, while its professional visualization and equipment manufacturer units brought in $601 million and $173 million, respectively. Nvidia’s auto and robotics segment remains small, at just 1 percent of overall sales. Yet its $586 million in revenue marked a 69 percent year-over-year jump, reflecting Nvidia’s push into “physical A.I.” “As a result of agentic A.I. and vision-language models, we are now seeing a breakthrough in physical A.I. in robotics and autonomous systems,” Huang told analysts.

    Nvidia Tops Wall Street Estimates, But China Tensions Cloud Its A.I. Chip Business

    Alexandra Tremayne-Pengelly

    Source link

  • Tesla upgrades EV voice assistant system with AI from DeepSeek and ByteDance

    Tesla is rolling out an upgraded voice assistant system for its electric vehicles (EVs) in mainland China, adopting artificial intelligence from DeepSeek and ByteDance to better engage with customers in the world’s largest automotive market.

    DeepSeek’s namesake chatbot would be used for “AI interaction”, which enables a Tesla EV’s driver to have casual conversations with the system, while also getting the latest news and weather information, according to the updated terms of use posted this month on the US carmaker’s mainland website.

    ByteDance’s Doubao large language model (LLM) would facilitate voice commands for navigation as well as in-vehicle media and amenities such as air conditioning, according to the updated terms.

    Do you have questions about the biggest topics and trends from around the world? Get the answers with SCMP Knowledge, our new platform of curated content with explainers, FAQs, analyses and infographics brought to you by our award-winning team.

    A user activates the upgraded voice assistant system by saying “Hey, Tesla” or another designated phrase, providing a more intuitive approach than clicking a button on either the EV’s steering wheel or multimedia terminal.

    Volcano Engine, the cloud computing services unit of ByteDance, is responsible for the AI systems integration using an encrypted application programming interface, a protocol that enables different software applications to communicate.

    Tesla did not immediately respond to a request for comment on Friday.

    Visitors check out a Tesla electric vehicle on display at the third China International Supply Chain Expo in Beijing on July 16, 2025. Photo: EPA alt=Visitors check out a Tesla electric vehicle on display at the third China International Supply Chain Expo in Beijing on July 16, 2025. Photo: EPA>

    Tesla’s latest initiative reflects the carmaker’s efforts to boost orders on the mainland’s highly competitive EV market, as the AI systems from DeepSeek and ByteDance would appeal to domestic buyers.

    Details on when Tesla’s upgraded voice assistant system would be available and on which models remain unknown. Tesla’s recently launched six-seat Model Y L SUV, which supports a voice wake-up feature, will start deliveries next month.

    Tesla’s updated terms, meanwhile, cautioned users that AI-generated content “may be incomplete, incorrect or contextually unsuitable”, adding that the technology should not be used to “endanger national security” or “disclose state secrets” as stipulated by China’s laws.

    Still, Tesla was nearly half a year late in adopting Chinese AI solutions. As of mid-February, more than a dozen domestic carmakers – including BYD, Geely and Stellantis-backed start-up Leapmotor – had already announced plans to release cars with DeepSeek-enabled AI features.

    Total EV deliveries – comprising passenger cars and commercial vehicles like buses – slid 5 per cent from a month earlier to 1.26 million units in July, according to data from the government-backed China Association of Automobile Manufacturers. It was the first month-on-month drop in the Chinese EV market since May.

    ByteDance has become a popular AI supplier for carmakers on the mainland. Last year, the TikTok and Douyin owner teamed up with Mercedes-Benz to integrate its LLM into the German carmaker’s in-car systems in China.

    The Beijing-based unicorn ByteDance had also formed an “automobile LLM ecosystem alliance” with more than 20 firms that included Geely and Great Wall Motor.

    This article originally appeared in the South China Morning Post (SCMP), the most authoritative voice reporting on China and Asia for more than a century. For more SCMP stories, please explore the SCMP app or visit the SCMP’s Facebook and Twitter pages. Copyright © 2025 South China Morning Post Publishers Ltd. All rights reserved.

    Copyright (c) 2025. South China Morning Post Publishers Ltd. All rights reserved.

    Source link