AI game-changer or overhyped? DeepSeek faces scrutiny over bold claims
The Hangzhou-based startup's announcement that it developed R1 at a fraction of the cost of Silicon Valley's latest models immediately called into question assumptions about the United States's dominance in AI and the sky-high market valuations of its top tech firms.
Some sceptics, however, have challenged DeepSeek's account of working on a shoestring budget, suggesting that the firm likely had access to more advanced chips and more funding than it has acknowledged.
'It's very much an open question whether DeepSeek's claims can be taken at face value. The AI community will be digging into them and we'll find out,' Pedro Domingos, professor emeritus of computer science and engineering at the University of Washington, told Al Jazeera.
'It's plausible to me that they can train a model with $6m,' Domingos added.
'But it's also quite possible that that's just the cost of fine-tuning and post-processing models that cost more, that DeepSeek couldn't have done it without building on more expensive models by others.'
In a research paper released last week, the DeepSeek development team said they had used 2,000 Nvidia H800 GPUs – a less advanced chip originally designed to comply with US export controls – and spent $5.6m to train R1's foundational model, V3.
OpenAI CEO Sam Altman has stated that it cost more than $100m to train its chatbot GPT-4, while analysts have estimated that the model used as many as 25,000 more advanced H100 GPUs.
The announcement by DeepSeek, founded in late 2023 by serial entrepreneur Liang Wenfeng, upended the widely held belief that companies seeking to be at the forefront of AI need to invest billions of dollars in data centres and large quantities of costly high-end chips.
It also raised questions about the effectiveness of Washington's efforts to constrain China's AI sector by banning exports of the most advanced chips.
Shares of California-based Nvidia, which holds a near-monopoly on the supply of GPUs that power generative AI, on Monday plunged 17 percent, wiping nearly $593bn off the chip giant's market value – a figure comparable with the gross domestic product (GDP) of Sweden.
While there is broad consensus DeepSeek's release of R1 at least represents a significant achievement, some prominent observers have cautioned against taking its claims at face value.
Palmer Luckey, the founder of virtual reality company Oculus VR, on Wednesday labelled DeepSeek's claimed budget as 'bogus' and accused too many 'useful idiots' of falling for 'Chinese propaganda'.
'It is pushed by a Chinese hedge fund to slow investment in American AI startups, service their own shorts against American titans like Nvidia, and hide sanction evasion,' Luckey said in a post on X.
'America is a fertile bed for psyops like this because our media apparatus hates our technology companies and wants to see President Trump fail.'
In an interview with CNBC last week, Alexandr Wang, CEO of Scale AI, also cast doubt on DeepSeek's account, saying it was his 'understanding' that it had access to 50,000 more advanced H100 chips that it could not talk about due to US export controls.
Wang did not provide evidence for his claim.
Tech billionaire Elon Musk, one of US President Donald Trump's closest confidants, backed DeepSeek's sceptics, writing 'Obviously' on X under a post about Wang's claim.
DeepSeek did not respond to requests for comment.
But Zihan Wang, a PhD candidate who worked on an earlier DeepSeek model, hit back at the startup's critics, saying, 'Talk is cheap.'
'It's easy to criticize,' Wang said on X in response to questions from Al Jazeera about the suggestion that DeepSeek's claims should not be taken at face value.
'If they'd spend more time working on the code and reproduce the DeepSeek idea theirselves it will be better than talking on the paper,' Wang said, using an English translation of a Chinese idiom about people who engage in idle talk.
He did not respond directly to a question about whether he believed DeepSeek had spent under $6m and used less advanced chips to train R1's foundational model.
In a 2023 interview with Chinese media outlet Waves, Liang said his company had stockpiled 10,000 of Nvidia's A100 chips – which are older than the H800 – before the administration of then-US President Joe Biden banned their export.
Users of R1 also point to limitations it faces due to its origins in China, namely its censoring of topics considered sensitive by Beijing, including the 1989 massacre in Tiananmen Square and the status of Taiwan.
In a sign that the initial panic about DeepSeek's potential impact on the US tech sector had begun to recede, Nvidia's stock price on Tuesday recovered nearly 9 percent.
The tech-heavy Nasdaq 100 rose 1.59 percent after dropping more than 3 percent the previous day.
Tim Miller, a professor specialising in AI at the University of Queensland, said it was difficult to say how much stock should be put in DeepSeek's claims.
'The model itself gives away a few details of how it works, but the costs of the main changes that they claim – that I understand – don't 'show up' in the model itself so much,' Miller told Al Jazeera.
Miller said he had not seen any 'alarm bells' but there are reasonable arguments both for and against trusting the research paper.
'The breakthrough is incredible – almost a 'too good to be true' style. The breakdown of costs is unclear,' Miller said.
On the other hand, he said, major breakthroughs do happen occasionally in computer science.
'These massive-scale models are a very recent phenomenon, so efficiencies are bound to be found,' Miller said.
'Given they knew that this would be reasonably straightforward for others to reproduce, they would have known that they would look stupid if they were bullsh***ing everyone. There is a team already committed to trying to reproduce the work.'
Lucas Hansen, co-founder of the nonprofit CivAI, said that while it was difficult to know whether DeepSeek circumvented US export controls, the startup's claimed training budget referred to V3, which is roughly equivalent to OpenAI's GPT-4, not R1 itself.
'GPT-4 finished training late 2022. There has been a lot of algorithmic and hardware improvements since 2022, driving down the cost of training a GPT-4 class model. A similar situation happened for GPT-2. At the time it was a serious undertaking to train, but now you can train it for $20 in 90 minutes,' Hansen told Al Jazeera.
'DeepSeek made R1 by taking a base model – in this case V3 – and applying some clever methods to teach that base model to think more carefully,' Hansen added.
'This teaching process is comparatively cheap when compared to the price of training the base model. Now that DeepSeek has published details about how to bootstrap a base model into a thinking model, we will see a huge number of new thinking models.'
Hashtags
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Qatar Tribune
38 minutes ago
- Qatar Tribune
Intel gets $2 bn lifeline in form of SoftBank equity investment
Agencies Intel will receive a $2 billion capital injection from Japanese technology giant SoftBank Group, marking a major show of confidence in the struggling U.S. chipmaker as it works through a major turnaround. The equity investment, announced by the two companies on Monday, is a lifeline for the once-iconic U.S. chipmaker which has struggled to compete after years of management blunders that left it with virtually no foothold in the booming artificial intelligence chip industry. It will make SoftBank a top-10 shareholder of Intel and add to the Japanese tech investor's ambitious bet on artificial intelligence that includes the $500 billion Stargate U.S. data center project. 'SoftBank's investment helps, but it is not what is going to move the dial for Intel,' said Amir Anvarzadeh, Japan equity strategist at Asymmetric Advisors. 'It's more to maintain this very good relationship he has with Trump,' he said, referring to SoftBank CEO Masayoshi Son. The deal follows media reports last week that the U.S. government may buy a stake in Intel, after a meeting between new CEO Lip-Bu Tan and President Donald Trump after the latter called for the chief executive to resign over his ties to Chinese firms. But after meeting with him, Trump relented, saying Tan had an 'amazing story.' It also comes as Tokyo pledged a $550 billion investment package into the U.S. last month as part of a trade deal with Washington. The Intel investment is not currently part of that package, a Japanese government source with knowledge of the negotiations said. SoftBank's decision to invest in Intel is not connected to Trump, a person familiar with the matter told Reuters. 'Semiconductors are the foundation of every industry,' Son said in a statement. 'This strategic investment reflects our belief that advanced semiconductor manufacturing and supply will further expand in the United States, with Intel playing a critical role.' SoftBank will pay $23 per Intel share, a slight discount to Monday's closing price of $23.66. The investment will come via a primary issuance of common stock by Intel, and, based on the U.S. company's market capitalization at close of trading on Monday, represent an equity stake of just under 2%, an Intel spokesperson said. The Japanese company would become the sixth largest investor in Intel, according to LSEG shares closed down 4% on Tuesday following the announcement, while Intel surged 5.6% in after-market hours trading. The Japanese company will only take an equity stake in Intel and will neither seek a board seat nor commit to buying Intel's chips, the person familiar with the matter said. SoftBank posted its first profit in four years in the April-June quarter as it raked in gains from its investment portfolios. It is a major shareholder in Arm Holdings, a British semiconductor and software design company. Intel has struggled financially and recorded an annual loss of $18.8 billion in 2024, its first such loss since 1986, as it grapples with multiple challenges. Its longtime rival AMD has been gaining share in Intel's mainstay personal computer and server semiconductor markets, while its ambitious and costly plan for a chip contracting business that rivals that of Taiwan's TSMC has failed to take company is now considering a significant change to its contract chip manufacturing business to win major customers, Reuters reported last month, in a potentially expensive shift from its previous strategies. 'Intel's dual role as designer and manufacturer/fabricator uniquely positions it as potentially the best platform in the U.S. to compete with TSMC,' said Charu Chanana, chief investment strategist at News reported earlier on Monday that the U.S. government is in talks to take a 10% stake in Intel. Tan, a chip industry veteran who also served as a SoftBank board member before quitting in 2022, thanked Son for 'the confidence he has placed in Intel with this investment.' The Intel funding is the latest in the Japanese company's run of mammoth investment announcements in 2025, which include committing $30 billion to ChatGPT maker OpenAI as well as leading the financing for Stargate.
Qatar Tribune
40 minutes ago
- Qatar Tribune
After years of sourcing struggles, United States dryer maker cuts China out
Agencies Denis Gagnon had long wanted to establish a manufacturing firm supplied only by American-made parts, a vision shaped by his experience in the international corporate world and his assessment of business risks offshore. He was an outlier in an era of globalisation when businesspeople, including his US peers, waxed lyrical about outsourcing from Asia to minimise production costs. The Gagnons, including his son William, run a medium-sized company called Excel Dryer in the US state of 2018, the year US President Donald Trump raised tariffs on Chinese imports in his first term, the father and son knew they had the right plan. Their conviction was reinforced earlier this year when the world's two largest economies escalated duties on each other's imports. 'The tariff increase only solidified our decision to move forward,' said William Gagnon, who is now executive vice-president of the company. After eight tough years searching for a motor supplier outside China, the maker of electric hand dryers achieved its goal of sourcing 100 per cent of its parts domestically by 2023, William Gagnon told the Post. Today, Excel Dryer does not have to worry about paying tariffs on imports from China. The move to source goods locally came as the Trump administration pushes to restore US manufacturing in industries ranging from shipbuilding and copper to rare earth metals. 'We are happy we don't have to worry about it,' William Gagnon said. 'We're not on the edge of our seats worrying about what's going to happen. We are able to focus on more important and strategic things.' s While sourcing most components in the US was simple enough, replacing the firm's Chinese motor suppliers turned into a slog for the family-owned company, he said. Excel Dryer evaluated at least a dozen American motor suppliers before selecting one, the executive VP added. 'It was difficult to find an American motor vendor that could match the size, performance and costs of motors made in China because of the volume and lower labour costs,' he said. 'That's why other companies, like our competitors, have moved away from domestic sourcing in favour of importing components or the whole product to stay cost-competitive.' Excel Dryer was looking for 'reliability and consistency' when the Gagnons set out to source all components from the US. 'However, US parts often come at a higher cost,' he said. 'Chinese suppliers dominate on price, but longer [order] lead times, inconsistent quality and growing geopolitical tensions add risk.' He called the search for a motor supplier 'very thorough', covering performance, durability, cost, production capacity and reliability. Although he did not disclose the costs involved, he described it as a 'significant effort' and not an 'easy process'. Excel now gets its dryer motors from Scott Fetzer Electrical Group, based in the US state of Tennessee. William Gagnon said the company allows Excel to communicate directly with its management and negotiate purchase deals. The American vendor ended up offering higher 'quality control', shorter order times and 'greater supply chain stability' than Chinese counterparts, he said. 'Keeping the product American-made is not easy. It's taken a lot of focus and effort by us.' Excel Dryer does not disclose what it pays for motors, but William Gagnon said that American and Chinese engines have about the same 'landed cost' when factoring in tariffs. Still, many Chinese analysts doubt whether American firms can follow suit and decouple from Asian suppliers. Charles Chang, a finance professor at Fudan University in Shanghai, pointed out that other American light industrial firms facing Trump's tariffs had tried to replace Chinese suppliers with American ones but usually ended up paying the duties anyway or raising companies are too 'fragile' to afford the higher-cost, harder-to-find American suppliers, he said. 'These stories are popular, the workaround concept, but it's a small percentage, a very small group of people.' Excel, founded in 1997, does not disclose revenue or output figures but calls itself a mid-sized company with more than 15 per cent growth in annual revenue. The executive VP said Excel would not raise prices on account of onshoring its dryer motors. Prices start at US$535 per unit on Amazon, compared to an average of US$400 – US$500 per unit, across brands, as estimated by the US-based online B2B retailer HomElectrical Electric Dryer has saved money by shipping hundreds of dryers to countries that lowered duties on US imports after April 2, when Trump raised tariffs worldwide, the executive VP said. Some countries managed to get the US to lower tariffs by easing their own trade barriers that had long irked American exporters.'It's helping us to export because [Trump tariffs] reduce barriers to export into those markets.'
Qatar Tribune
41 minutes ago
- Qatar Tribune
Passwords under threat as tech giants seek tougher security
Agencies Fingerprints, access keys and facial recognition are putting a new squeeze on passwords as the traditional computer security method -- but also running into public hesitancy. 'The password era is ending,' two senior figures at Microsoft wrote in a July blog post. The tech giant has been building 'more secure' alternatives to log in for years -- and has since May been offering them by default to new users. Many other online services -- such as artificial intelligence giant OpenAI's ChatGPT chatbot -- require steps like entering a numerical code emailed to a user's known address before granting access to potentially sensitive data. 'Passwords are often weak and people re-use them' across different online services, said Benoit Grunemwald, a cybersecurity expert with Eset. Sophisticated attackers can crack a word of eight characters or fewer within minutes or even seconds, he pointed out. And passwords are often the prize booty in data leaks from online platforms, in cases where 'they are improperly stored by the people supposed to protect them and keep them safe,' Grunemwald said. One massive database of around 16 billion login credentials amassed from hacked files was discovered in June by researchers from media outlet Cybernews. The pressure on passwords has tech giants rushing to find safer alternatives. One group, the Fast Identity Online Alliance (FIDO) brings together heavyweights including Google, Microsoft, Apple, Amazon and TikTok. The companies have been working on creating and popularizing password-free login methods, especially promoting the use of so-called access keys. These use a separate device like a smartphone to authorize logins, relying on a pin code or biometric input such as a fingerprint reader or face recognition instead of a password. Troy Hunt, whose website Have I Been Pwned allows people to check whether their login details have been leaked online, says the new systems have big advantages. 'With passkeys, you cannot accidentally give your passkey to a phishing site' -- a page that mimics the appearance of a provider such as an employer or bank to dupe people into entering their login details -- he said. But the Australian cybersecurity expert recalled that the last rites have been read for passwords many times before. 'Ten years ago we had the same question... the reality is that we have more passwords now than we ever did before,' Hunt said. Although many large platforms are stepping up login security, large numbers of sites still use simple usernames and passwords as credentials. The transition to an unfamiliar system can also be confusing for have to be set up on a device before they can be used to log in. Restoring them if a PIN code is forgotten or trusted smartphone lost or stolen is also more complicated than a familiar password reset procedure. 'The thing that passwords have going for them, and the reason that we still have them, is that everybody knows how to use them,' Hunt said. Ultimately the human factor will remain at the heart of computer security, Eset's Grunemwald said. 'People will have to take good care of security on their smartphone and devices, because they'll be the things most targeted' in future, he warned.
