The real hype behind DeepSeek, according to AI experts
DeepSeek turned the tech world on its head last month – and for good reason, according to artificial intelligence experts, who say we're likely only seeing the beginning of the Chinese tech startup's influence on the AI field.
DeepSeek grabbed headlines in late January with its R1 AI model, which the company says can roughly match the performance of Open AI's o1 model at a fraction of the cost. Tech stocks tumbled as DeepSeek briefly unseated ChatGPT to become the top app in Apple's App Store.
The achievement pushed US tech behemoths to question America's standing in the AI race against China – and the billions of dollars behind those efforts. While Vice President JD Vance didn't mention DeepSeek or China by name in his remarks at the Artificial Intelligence Action Summit in Paris on Tuesday, he certainly emphasized how big of a priority it is for the United States to lead the sector.
'The United States of America is the leader in AI, and our administration plans to keep it that way,' he said, although he added that 'America wants to partner' with other countries.
But it's not just DeepSeek's efficiency and power. The way DeepSeek R1 can reason and 'think' through answers to provide quality results, along with the company's decision to make key parts of its technology publicly available, will also push the field forward, experts say.
While AI has long been used in tech products, it's reached a flashpoint over the last two years thanks to the rise of ChatGPT and other generative AI services that have reshaped the way people work, communicate and find information. It's made Wall Street darlings out of companies like chipmaker Nvidia and upended the trajectory of Silicon Valley giants. So any development that can help build more capable and efficient models is sure to be closely watched.
'This is definitely not hype,' said Oren Etzioni, former CEO of the Allen Institute for Artificial Intelligence. 'But also, this is a very fast-moving world.'
Tech leaders have been quick to respond to DeepSeek's rise. Google DeepMind CEO Demis Hassabis called the hype around DeepSeek 'exaggerated,' but also said its model as 'probably the best work I've seen come out of China,' according to CNBC.
Microsoft CEO Satya Nadella said on the company's quarterly earnings call in January that DeepSeek has some 'real innovations,' while Apple CEO Tim Cook said on the iPhone maker's earnings call that 'innovation that drives efficiency is a good thing.'
But the attention hasn't all been positive. Semiconductor researcher SemiAnalysis cast doubt over DeepSeek's claims that it only cost $5.6 million to train. OpenAI told The Financial Times it found evidence that DeepSeek used the US company's models to train its own competitor.
'We are aware of and reviewing indications that DeepSeek may have inappropriately distilled our models, and will share information as we know more,' an OpenAI spokesperson said in a comment to CNN. DeepSeek could not immediately be reached for comment.
And a pair of US lawmakers has already called for the app to be banned from government devices after security researchers highlighted its potential links to the Chinese government, as the Associated Press and ABC News reported. Similar concerns have been raised about the popular social media app TikTok, which must be sold to an American owner or risk being banned in the US.
'DeepSeek is the TikTok of (large language models),' Etzioni said.
Tech giants are already thinking about how DeepSeek's technology can influence their products and services.
'What DeepSeek gave us was essentially the recipe in the form of a tech report, but they didn't give us the extra missing parts,' said Lewis Tunstall, a senior research scientist at Hugging Face, an AI platform that offers tools for developers.
Tunstall is leading an effort at Hugging Face to fully open source DeepSeek's R1 model; while DeepSeek provided a research paper and the model's parameters, it didn't reveal the code or training data.
Nadella said on Microsoft's earnings call that Windows Copilot+ PCs, or PCs built to a certain spec to support AI models, will be able to run AI models distilled from DeepSeek R1 locally. Mobile chipmaker Qualcomm said on Tuesday that models distilled from DeepSeek R1 were running on smartphones and PCs powered by its chips within a week.
AI researchers, academics and developers are still exploring what DeepSeek means for the advancement of AI.
DeepSeek's model isn't the only open-source one, nor is it the first to be able to reason over answers before responding; OpenAI's o1 model from last year can do that, too.
What makes DeepSeek significant is the way it can reason and learn from other models, along with the fact that the AI community can see what's happening behind the scenes. Those who use the R1 model in DeepSeek's app can also see its 'thought' process as it answers questions.
'You can see the wheels turning inside the machine,' Durga Malladi, senior vice president and general manager for technology planning and edge solutions at Qualcomm, said to CNN.
Tunstall thinks we may see a wave of new models that can reason like DeepSeek in the not-too-distant future. That could be critical as tech giants race to build AI agents, which Silicon Valley generally believes are the next evolution of the chatbot and how consumers will interact with devices – although that shift hasn't quite happened yet.
Grok 3, the next iteration of the chatbot on the social media platform X, will have 'very powerful reasoning capabilities,' its owner, Elon Musk, said on Thursday in a video appearance during the World Governments Summit.
For now, the AI community will keep tinkering with what DeepSeek has to offer. That is, until the next breakthrough comes along.
'I certainly predict that in the next 12 months, it'll be supplanted by something else,' said Etzioni. 'But it's a very real advance.'
Hashtags
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Forbes
39 minutes ago
- Forbes
The AI Era Enters Its Sovereign Phase
Generative AI adoption started in late 2022 with public adoption of models like ChatGPT and Llama. As it drives towards its next phase of value creation with reasoning, also referred to as agentic AI, it has recently crossed the boundary from a consumer-centric application into an enterprise application. Right on the heels of this adoption is also another phase of value creation – Sovereign AI. What Is Sovereign AI? Sovereign AI refers to artificial intelligence that is developed, maintained, and controlled within a specific nation's or organization's jurisdiction, ensuring independence from external influences. This artificial intelligence is designed to align with local regulations, ethical standards, and strategic priorities, allowing governments and enterprises to maintain autonomy over their AI-driven operations. The Opportunity To Reign Supreme (Or At Least Be At The Front Of The Pack) Nvidia CEO Jensen Huang recently stated that 'AI is now an essential form of national infrastructure – just like energy, telecommunications and the internet.' Indeed, many leading countries such as the United States, United Kingdom, China, France, Denmark and the United Arab Emirates have launched sovereign AI initiatives. Stargate is an example of such an initiative from the United States. Additionally, leading AI enablers like Nvidia and OpenAI, have initiatives targeted specifically at helping entities establish their own sovereign AI capabilities. Sovereign AI is particularly crucial in areas like national security, defense, and critical infrastructure, where reliance on foreign AI models could pose risks related to data privacy, cybersecurity, or geopolitical dependencies. By building and maintaining custom AI capabilities, nations and organizations can safeguard their technological sovereignty while fostering innovation tailored to their unique needs. Moving Forward With Sovereign AI While this is a gross oversimplification of how complicated this task is for national leaders to undertake, the following are some critical areas that must be addressed in embarking on the sovereign AI journey: To this end, AI enablers like Nvidia and leading countries such as France have started to organize events. For example, at the upcoming Viva Technology event in Paris this coming June, Jensen Huang and Nvidia have organized a dedicated GTC event where interested parties can learn more. As mentioned earlier, it is important to keep in mind that sovereign AI isn't necessarily limited to national entities. Any sufficiently capable entity, whether they be nations, companies, organizations or universities interested in securing their own AI systems and capabilities from data curation and model creation to specified and focused outcomes can take advantage of sovereign AI.
Yahoo
an hour ago
- Yahoo
Labour is cosying up to China after years of rollercoaster relations
The sprawling city of Chongqing in southwestern China is an incredible sight. Built on mountainous terrain and crisscrossed by rivers, it is connected by vast elevated roads. Trains even run through some buildings. TikTokers have begun documenting their commutes in the striking urban architecture, generating millions of likes and much hype. But it is also where, on a somewhat quieter trip, mayors and their deputies from the UK recently visited - the largest British civic delegation to visit the country in modern history. The whole trip, which took place in March, received substantial Chinese media coverage, despite flying more under the radar in the UK. The impression it left on some of the politicians who travelled there was vast. "[The city is] what happens if you take the planning department and just say 'yes' to everything," reflects Howard Dawber, deputy London mayor for business. "It's just amazing." The group travelled to southern Chinese cities, spoke to Chinese mayors and met Chinese tech giants. So impressed was one deputy mayor that, on returning home, they bought a mobile phone from Chinese brand Honor (a stark contrast from the days the UK banned Huawei technology from its 5G networks, just a few years ago). Roughly half-a-dozen deals were signed on the back of the trip. The West Midlands, for example, agreed to establish a new UK headquarters in Birmingham for Chinese energy company EcoFlow. But the visit was as much about diplomacy as it was trade, says East Midlands deputy mayor Nadine Peatfield, who attended. "There was a real hunger and appetite to rekindle those relationships." To some, it was reminiscent of the "golden era" of UK-China relations, a time when then-Prime Minister David Cameron and Chinese President Xi Jinping shared a basket of fish and chips and a pint. Those days have long felt far away. Political ties with China deteriorated under former UK Conservative Prime Ministers Boris Johnson, Rishi Sunak and Liz Truss. The last UK prime minister to visit China was Theresa May, in 2018. But the recent delegation - and the talk of Sir Keir Starmer possibly visiting China later this year - suggests a turning point in relations. But to what greater intent? The course correction seemed to begin with the closed-door meeting between Sir Keir and Chinese President Xi in Brazil last November. The prime minister signalled that Britain would look to cooperate with China on climate change and business. Since then, Labour's cautious pursuit of China has primarily focused on the potential financial upsides. In January, Chancellor Rachel Reeves co-chaired the first UK-China economic summit since 2019, in Beijing. Defending her trip, she said: "Choosing not to engage with China is no choice at all." Reeves claimed re-engagement with China could boost the UK economy by £1bn, with agreements worth £600m to the UK over the next five years — partially achieved through lifting barriers that restrict exports to China. Soon after, Energy Secretary Ed Miliband resumed formal climate talks with China. Miliband said it would be "negligence" to future generations not to have dialogue with the country, given it is the world's biggest carbon emitter. Labour simply describes its approach as "grown-up". But it all appears to be a marked shift from the last decade of UK-China relations. During the so-called "golden era", from 2010, the UK's policy towards China was dominated by the Treasury, focusing on economic opportunities and appearing to cast almost all other issues, including human rights or security, aside. By September 2023, however, Rishi Sunak said he was "acutely aware of the particular threat to our open and democratic way of life" posed by China. Labour claimed in its manifesto that it would bring a "long-term and strategic approach". China has a near monopoly on extracting and refining rare earth minerals, which are critical to manufacturing many high-tech and green products. For example, car batteries are often reliant on lithium, while indium is a rare metal used for touch screens. This makes China a vital link in global supply chains. "China's influence is likely to continue to grow substantially globally, especially with the US starting to turn inwards," says Dr William Matthews, a China specialist at Chatham House think tank. "The world will become more Chinese, and whilst that is difficult for any Western government, there needs to be sensible engagement from the get-go." Andrew Cainey, a director of the UK National Committee on China, an educational non-profit organisation, says: "China has changed a lot since the Covid-19 pandemic. To have elected officials not having seen it, it's a no brainer for them to get back on the ground". Certainly many in the UK's China-watching community believe that contact is an essential condition to gain a clearer-eyed view of the opportunities posed by China, but also the challenges. The opportunities, some experts say, are largely economic, climate and education-related. Or as Kerry Brown, Professor of Chinese Studies at King's College London, puts it: "China is producing information, analysis and ways of doing things that we can learn from". He points to the intellectual, technological, AI, and life sciences opportunities. Not engaging with China would be to ignore the realities of geopolitics in the 21st century, in Dr Matthew's view, given that it is the world's second largest economy. However he also believes that engagement comes with certain risks. But Charles Parton, who spent 22 years of his diplomatic career working in or on China, raises questions about the UK's economic and national security. For example, the government is reportedly weighing up proposals for a Chinese company to supply wind turbines for an offshore windfarm in the North Sea. Mr Parton warns against allowing China access to the national grid: "It wouldn't be difficult in a time of high tension to say, 'by the way, we can turn off all your wind farms'". But earlier this year, the China Chamber of Commerce to the EU issued a statement expressing concern over the "politicisation" of deals between wind developers in Europe and Chinese turbine suppliers. Xi's real test is not Trump's trade war North and South Korea are in an underground war - Kim Jong Un might now be winning The Conservative Party faces problems - is its leader one of them? James Sullivan, director of Cyber and Tech at defence think tank Rusi, notes there are also some questions around cyberspace. "China's activities in cyberspace appear to be more strategically and politically focused compared to previous opportunistic activities," he says. As for defence, the UK's recently published defence review describes China as a "sophisticated and persistent challenge", with Chinese technology and its proliferation to other countries "already a leading challenge for the UK". Ken McCallum, MI5 director general, meanwhile, has previously warned of a sustained campaign on an "epic scale" of Chinese espionage abroad. But Prof Brown pushes back on some concerns about espionage, saying some media narratives about this are a "fairytale". Beijing has always dismissed accusations of espionage as attempts to "smear" China. Sir Keir and his team will no doubt be closely monitoring how this is all viewed by Washington DC. Last month, President Donald Trump's trade advisor Peter Navarro described Britain as "an all too compliant servant of Communist China", urging the UK against deepening economic ties. "When it comes to foreign policy towards China, America's influence on policy will be quite substantive compared with say continental Europe," says Dr Yu Jie of China Foresight at LSE IDEAS think tank. Most analysts I speak to in both the UK and China are still clear on the need for the two countries to get back in the same room, even if they differ on where to draw the line: in which areas should Westminster cooperate and where should it stay clear. These red lines have not yet been drawn, and experts say that without some kind of playbook, it is difficult for businesses and elected officials to know how to engage. "You can only keep firefighting specific issues for so long without developing a systematic plan," warns Mr Cainey. Certain thorny issues have arisen, including Chinese investments in the UK. For example in April when the government seized control of British Steel from its former Chinese owner Jingye, to prevent it from being closed down, Business Secretary Jonathan Reynolds admitted that he would "look at a Chinese firm in a different way" when considering investment in the UK steel industry. China's foreign ministry spokesperson, Lin Jian, warned that Labour should avoid "linking it to security issues, so as not to impact the confidence of Chinese enterprises in going to the UK". After Starmer met Xi last year, he said the government's approach would be "rooted in the national interests of the UK", but acknowledged areas of disagreement with China, including on human rights, Taiwan and Russia's war in Ukraine. Securing the release of pro-democracy activist and British citizen Jimmy Lai from a Hong Kong prison is, he has said, a "priority" for the government. Labour's manifesto broadly pledged: "We will cooperate where we can, compete where we need to, and challenge where we must." What is still lacking, however, is the fine print. Asked about the British government's longer-term strategy, Mr Parton replied: "No.10 doesn't have a strategy." He tells me he has some specific advice: "Go with your eyes open," he says. "But have a clear idea of what needs protecting, and a willingness to take some short-term financial hits to protect long-term national security." Labour has suggested that some clarity on their approach will be provided through the delayed China "audit", a cross-government exercise launched last year, which will review the UK's relations with China. The audit is due to be published this month, but many doubt that it will resolve matters. "If we see a visit from Starmer to Beijing, that will be an indication that the two sides have actually agreed with something, and that they would like to change and improve their bilateral relationship," says Dr Yu. But many people in Westminster remain China-sceptic. And even if the audit helps Britain better define what it wants out of its relationship with China, the question remains, do MPs and businesses have the China-related expertise to get the best out of it? According to Ruby Osman, China analyst at the Tony Blair Institute, there is an urgent need to build the UK's China capabilities in a more holistic way, focusing on diversifying the UK's points of contact with China. "If we want to be in a position where we are not just listening to what Beijing and Washington want, there needs to be investment in the talent pipeline coming into government, but also think tanks and businesses who work with China," she argues. And if that's the case, then irrespective of whether closer ties with China is viewed as a security threat, an economic opportunity, or something in between, the UK might be in a better position to engage with the country. Top image credit: PA BBC InDepth is the home on the website and app for the best analysis, with fresh perspectives that challenge assumptions and deep reporting on the biggest issues of the day. And we showcase thought-provoking content from across BBC Sounds and iPlayer too. You can send us your feedback on the InDepth section by clicking on the button below.
Yahoo
an hour ago
- Yahoo
What is a GPT?
When you buy through links on our articles, Future and its syndication partners may earn a commission. The introduction of generative pre-trained transformers (GPTs) marked a significant milestone in the adoption and utility of artificial intelligence in the real world. The technology was created by the then fledgling research lab OpenAI, based on previous research done on transformers in 2017 by Google Labs. It was Google's white paper "Attention is all you need", which laid the foundation for OpenAI's work on the GPT concept. As seen in > Model matchup surprise > ChatGPT announcements > Goodbye ChatGPT-4 > Why ChatGPT 4.1 is a big deal Transformers provided AI scientists with an innovative method of taking user input, and converting it to something that could be used by the neural network using an attention mechanism to identify important parts of the data. This architecture also allows for the information to be processed in parallel rather than sequentially as with traditional neural networks. This provides a huge and critical improvement in speed and efficiency of AI processing. OpenAI's GPT architecture was released in 2018 with GPT-1. By significantly refining Google's transformer ideas, the GPT model demonstrated that large-scale unsupervised learning could produce an extremely capable text generation model which operated at vastly improved speeds. GPT's also uprated the neural networks' understanding of context which improved accuracy and provided human-like coherence. Before GPT, AI language models relied on rule-based systems or simpler neural networks like recurrent neural networks (RNNs), which struggled with long-range dependencies and contextual understanding. The story of the GPT architecture is one of constant incremental improvements ever year since launch. GPT-2 in 2019 introduced a model with 1.5 billion parameters, which started to provide the kind of fluent text responses where AI users are now familiar with. However it was the introduction of GPT-3 (and subsequently 3.5) in 2020 which was the real game-changer. It featured 175 billion parameters, and suddenly a single AI model could cope with a vast array of applications from creative writing to code generation. GPT technology went viral in November of 2022 with the launch of ChatGPT. Based on GPT 3.5 and later GPT-4, this astonishing technology instantly propelled AI into public consciousness in a massive way. Unlike previous GPT models, ChatGPT was fine-tuned for conversational interaction. Suddenly business users and ordinary citizens could use an AI for things like customer service, online tutoring or technical support. So powerful was this idea, that the product attracted a 100 million users in a mere 60 days. Today GPT is one of the top two AI system architectures in the world (along with Google's Gemini). Recent improvements have included multimodal capabilities, i.e. the ability to process not just text but also images, video and audio. OpenAI has also updated the platform to improve pattern recognition and enhance unsupervised learning, as well as adding agentic functionality via semi-autonomous tasks. On the commercial front, GPT powered applications are now deeply embedded in many different business and industry enterprises. Salesforce has Einstein GPT to deliver CRM functionality, Microsoft's Copilot is an AI assisted coding tool which incorporates Office suite automation, and there are multiple healthcare AI models which are fine-tuned to provide GPT powered diagnosis, patient interaction and medical research. At the time of writing the only two significant rivals to the GPT architecture are Google's Gemini system and the work being done by DeepSeek, Anthropic's Claude and Meta with its Llama models. The latter products also use transformers, but in a subtly different way to GPT. Google however is a dark horse in the race, as it's becoming clear that the Gemini platform has the potential to dominate the global AI arena within a few short years. Despite the competition, OpenAI remains firmly at the top of many leaderboards in terms of AI performance and benchmarks. Its growing range of reasoning models such as o1 and o3, and its superlative image generation product, GPT Image-1 which uses the technology, continue to demonstrate that there is significant life left in the architecture, waiting to be exploited.
