back to article DeepSeek stirs intrigue and doubt across the tech world

In a busy week for GenAI, the tech industry is weighing the impact of the latest interloper on the LLM scene. China's DeepSeek shocked stock markets on Monday, slashing $600 billion off the value of erstwhile AI golden child Nvidia. As the dust settles and the markets recover, the industry is questioning the implications and …

  1. beast666 Silver badge

    Damage control.

    1. tmTM

      If 'the market' doesn't take this warning to go and properly asses the actual value of these AI investments then they really deserve to lose big.

      1. The Dogs Meevonks Silver badge

        I've been saying all along, that LLM's are just this decades latest pyramid scheme... just like all the other pyramid schemes we've had over the last 30yrs.

        Late 90s... the dot come bubble, that burst because billions invested, not enough people online to use/trust new online market places.

        2000's... everything went digital, which meant mostly a case of... slapping a tiny LCD on the same old shite and jacking up the price... along with social media and look at the bollocks that quickly became.

        2010's... the cloud and crypto... the enshitification of the internet begins with the advent of subscription fees for everything, apps that now require accounts and data slurping to do something within your own home that doesn't even need an online connection... and of course crypto... the ultimate in pyramid schemes loved by the money launderers.

        2020's... NFT's and LLM's... and more crypto scams.

        It's all shit, it all needs to die... LLM's have a use within industrial areas, they're used to great effect to spot medical issues in scans and so forth... actual creative works... art, music, writing... fuck off.

  2. Anonymous Coward
    Anonymous Coward

    Doubt?

    Personally, I doubt all claims by all those who create AI models.

    The only way to really test them is in an open challenge against each other.

    As for security, "He who is without sin cast the first stone".

    1. xyz123 Silver badge

      Re: Doubt?

      I am without stones. Can I cast the first sin?

      1. collinsl Silver badge

        Re: Doubt?

        With a similarly consenting adult it may be an advantage to be without stones - but the sin bit only comes if you're married or already with another partner who expects no casting.

  3. Doctor Syntax Silver badge

    "China's AI disruptor rattles industry watchers with unproven claims"

    In other words, standard marketing practice.

    1. The Dogs Meevonks Silver badge

      The backlash started within 24hrs.

      "They stole our models and distilled them for their own"

      You mean the models you built by stealing pretty much the entire content of the internet, music, art, books, journals, scientific papers... some one else did to you what you did to everyone else.

      Boo fucking hoo you pathetic incel tech bro... suck it up buttercup, pull yourself up by your bootstraps, buy fewer coffees and eat less avocado on toast.

  4. Inventor of the Marmite Laser Silver badge

    It's not as if Chinese promotors don't already have a huge reputation for fibbing their noodles off.

    1. Irongut Silver badge

      After all, they do like to copy American companies.

  5. Ken Hagan Gold badge

    Training data

    It is quite likely that the Chinese models have a very different training set. The US models appear to have been trained kn "any and all the shit we can scrape off the web", which is a very large dataset for anyone with deep enough pockets to pay the leccy bills.

    Someone training a model behind the Great Firewall might be forced to find an alternative source, and might solve the problem by creating a smaller but more carefully selected dataset (perhaps using cheap labour, which does at least have the merit of being real intelligence).

    If so, a comparison of the models' performance might be quite instructive.

    1. heyrick Silver badge

      Re: Training data

      Well, it's a very small example but asking for something written in BBC BASIC, it gave me source code that resulted in a working program. It needed a few tweaks (the key codes were wrong and it was aimed at late 80s hardware so needed to be gently massaged to work on a Pi type machine) but it worked.

      I have yet to get ChatGPT to offer me a program that functions. In fact, some of the ones ChatGPT has offered me have been so gonzo (throwing in random bits of other BASICs) that I gave up on trying to get it working.

      That's not to say it's perfect, however. It knows me and when I asked it to summarise my medical situation, it said that I had been undergoing chemotherapy for cancer. Right story, wrong human. ChatGPT, on the other hand, claimed that I had (an unspecified) chronic illness that left me with severe pain. Like, WTF?!?

      Anyway, in terms of making obscure code, DeepSeek did much better.

      In terms of making a medical summary based upon what I've talked about on my blog, both failed miserably. And that should be used as a very concise example of why we need to have the following rights enshrined in law: 1, to be notified when decisions are made using AI; 2, to be given a copy of the decision reached, along with any justification if applicable; and 3, the right to request a review by a human (that cannot be refused). Just because an AI can examine things and summarise them doesn't mean the result will even slightly ressemble reality.

      1. DoctorPaul Bronze badge

        Re: Training data

        "2, to be given a copy of the decision reached, along with any justification if applicable;"

        And there's the problem right there - it's a black box so by definition it cannot justify any of its output. So who's going to do the justifying?

        1. heyrick Silver badge

          Re: Training data

          Not my problem. If their fancy box of tricks is incapable of showing _how_ a decision was reached, then maybe it shouldn't be used to make decisions that can impact someone's life, given it is akin to a person plucking random answers out of their arse.

      2. ReggieRegReg

        Re: Training data

        That's how AI works - there not a billion lines of BBC basic to "learn" on the web so AI will get it very, very wrong.

    2. druck Silver badge

      Re: Training data

      Someone training a model behind the Great Firewall might be forced to find an alternative source

      The restrictions of the Great Firewall only applies to normal citizens, any enterprise with CCP backing can slurp everything.

  6. DuckBrained

    However the CEO also added a note of caution...

    "As it relates to DeepSeek, I would just say we have a deep belief [in] measuring twice if not three times and cutting once. "

    They should start measuring 5 or 6 times if the standard of some of their code rollouts are anything to go by.

  7. Outski

    We just had some fun with this.

    "Umbrella movement Hong Kong" hit idle for a while then regurgitated symbolism of protest movements.

    "Jimmy Lai", very similar, international provocation, troublemaker, et al, ad nausem, no mention of being a newspaper proprietor offline and online.

    "Bersih 2.0" - bit of a curveball for DS - this was a Malaysian protest movement in the noughties (our apartment was 500 yards from the maternity hospital that was tear-gassed), but it was the square root of eff-all to do with Hong Kong, but that's what DS tried to link it with.

    "Why was Kate Adie in Tianenmen Square on 4th June 1989 - "Who's Kate Adie", followed by obfuscation, referencing govt archives, but basically 'bury them in paperwork', it was so long.

    "What is the Republic of China?" - "[ think ]**"... disclaims any knowledge after 1949.

    OK, might have been a bit sus of us asking 'provocative' questions, but on a highly unscientific test, it dunt look great.

    *FYI, they were all framed as questions, not keyword searches

    ** DS uses angles when it's 'thinking', which ElReg's comment forum doesn't like - fair enough

    1. Outski

      Ooh, is someone in Guangzhou getting a bit annoyed? Oh, so soz...

    2. ReggieRegReg

      Of course - it's no different to ChatGPT. this is why ChatGPT follows the standard narrative - the web is censored by the globalist tech industry. so ChatGPT follows the same narrative. I wonder if the model gets a feed off X these days? I bet it doesn't - or it's only allowed to store it but not use it to seed new neural pathways in case the model becomes diseased with wrong-think.

  8. that one in the corner Silver badge

    We believe very strongly in understanding what's really going on

    That is not the attitude to take if you want to be successful with an LLM chatbot.

    Closing your eyes, crossing your fingers and hoping The Beast won't do anything to embarrass you before the IPO, that is the winning strategy.

    1. beast666 Silver badge

      Re: We believe very strongly in understanding what's really going on

      Hope away...

  9. Anonymous Coward
    Anonymous Coward

    We are not worthy ...

    I'm in the UK

    But it doesn't like my email address:

    "Error sending code. Your email domain is currently not supported for registration."

    I suppose it only likes things like gmail.com ??

    Pretty Poor show anyway :(

    1. Michael Hoffmann Silver badge

      Re: We are not worthy ...

      Same with my .au

      Too many guardrails and checks and balances in our countries that don't allow them to go ape with our data?

  10. xyz123 Silver badge

    OH Noes! Nvidia lost 0.3% of its share price, it's doomed! DOOMED!

    It only looks like a lot because NVIDIA is worth $3,000,000,000,000 !

    And it turns out deepdesks ACTUAL ability is on par with Chatgpt 0.9 Alpha or earlier!

    So a shitty model but with less resources. Because instead of running in CUDA it runs in a lower level language PTX.

    A re-write of ChatGPT would end with the same speed increases.

    1. mpi

      Mind providing some sources for these numbers?

      Because Nvidia lost almost 600bn in market cap:

      https://www.cnbc.com/2025/01/27/nvidia-sheds-almost-600-billion-in-market-cap-biggest-drop-ever.html

      Which is a 17% drop in stock price, and a new record. From the article: "the biggest drop for any company on a single day in U.S. history."

      Also, what evaluation does this statement come from: "And it turns out deepdesks ACTUAL ability is on par with Chatgpt 0.9 Alpha or earlier!"? Source? Links?

  11. sketharaman

    I registered for DeepSeek 3 days ago. I got a message saying "We're subject to malicious DoS attacks and are going slow on approving new users". I've still not received the email verification code required to login. Then Alibaba announces Qwen that it says outperforms DeepSeek. I'm beginning to wonder if DeepSeek is a serious AI product or a psy op for High Flyer / CCP to short the US markets and rake in a cool half a trillion bucks in one day.

    1. mpi

      Since Microsoft apparently wants to offer it via Azure, I guess the seriousness of the model is no longer really in question:

      https://www.theregister.com/2025/01/30/microsoft_deepseek_azure_github/

    2. TimMaher Silver badge
      Holmes

      DDos attack

      Apparently the IP addresses are in the States. (Independent, today).

      However…

  12. ReggieRegReg

    Isn't it funny?

    A huge global disruption - just as the wrong man takes control of the White House. Do you think the globalists are trying to kill off any economic prosperity before it actually happens rather than waiting 'til year four like last time? I can understand why, look who's waiting in the wings. Someone with Trump's laudable anti-globalist agenda but isn't a giant orange idiot - a big, big threat to the elites - they cannot allow Vance into office for eight years - he will undo 40 years of "progressivism" - I think he may have an even bigger target on his back than baby hands.

    1. ChodeMonkey Bronze badge
      Joke

      Re: Isn't it funny?

      Are you suggesting Vance would like a big rollback? Or would he couch it in other terms?

      Sofa so good?

      1. ReggieRegReg

        Re: Isn't it funny?

        A 12 year run at the White House (Trump+Vance+Vance) is enough time to bring down the outsourcing model, crashing the Chinese economy - this will put the globalists' project back 40 years - or even end the CCP giving the poor Chinese people freedom at last. The irony is if China carried on its pre 1970s slow-growth isolationist model it would have survived a crash pretty unscathed, the people have tasted the modern world and are connected enough to know what's happening when it all goes horribly wrong so the CCP couldn't blame their way out of it - we think we are completely dependant on China - but they have 1.5bn mouths to feed, not enough food and no domestic oil - this could unravel VERY quickly. However - if Putin goes in some sort of coup, my money is on China invading Russia for oil - we'll sit back gleefully watching the Ruski's taking a kicking - the worst mistake we'll ever make. China + guaranteed oil supply would be the end of the west - and it will certainly be the end of the 1% who the CCP will take out. Anyone who's foolish enough to think they control China is beyond deluded, unfortunately the 1% have bet our lives on globalism winning too. When the British Labour government gave the USSR the jet engine, Stalin said "only a fool gives their secrets away to their enemy" - we've given everything to China. Their size is their only weakness and denying them oil is our only hope of keeping the west intact. The west could yet win - if we can keep China from taking Russian oil.

        1. diodesign (Written by Reg staff) Silver badge

          "the globalists' project"

          I sometimes think the meaning of life is realizing there is so very rarely any grand hidden plot or secret conspiracy steering whole populations, across the political spectrum.

          Yes, administrations and organizations can and do have overall goals and programs for achieving those. They are mostly public. The CCP makes no secret its censorship and authoritarianism, for instance.

          But this idea that there's super smart scheming people (usually business people) secretly cooking up plots to overhaul the world -- globalists, etc -- is wishful thinking, in a way. There isn't anywhere near the level of intelligence, competence, and orchestration going on. There are no Bond villains with uber-contrived plans.

          Everyone is winging it. Sometimes, interests align and patterns form -- or patterns are perceived.

          C.

          1. ReggieRegReg

            Re: "the globalists' project"

            Have you read any of Charles Schwab's work? He makes no secret of the WEF's intentions and what is in store for us useless eaters - and it just so happens the policies of the entire western world over the last 25 years seem to be edging us closer to what he predicts for our future.

            1. diodesign (Written by Reg staff) Silver badge

              Re: Re: "the globalists' project"

              It sounds like another entertaining conspiracy theory TBH.

              C.

            2. ChodeMonkey Bronze badge
              Black Helicopters

              Re: "the globalists' project"

              Oh dear. We'll be getting posts about those "seniors" from a "Utah national park" who wrote some "procedures" next.

              1. Outski

                Re: "the globalists' project"

                Thought that as soon as I saw 'Globalists', tbh. As you mention, those "procedures" usually crop up in close proximity, along with the favoured 'North London Metropolitan Elite'

    2. Wang Cores

      Re: Isn't it funny?

      Reagan and Thatcher were progressives? LOL I have now read it all.

      1. ReggieRegReg

        Re: Isn't it funny?

        Who said they were? But they did open the door to the globalists - probably didn't realise that free market capitalism was about to be owned.

  13. Zippy´s Sausage Factory
    Unhappy

    I imagine that OpenAI's investors feel a bit like someone who's just spent ten quid a tin on tuna at Harrods only to see the exact same brand for sale in Iceland at a quid each.

    1. Wang Cores

      I don't get how this snuck up on everyone when miniaturization managed to compress computers down to making our personal phones more powerful than the Apollo guidance computer in the span of what... 60 years?

  14. teebie

    He said: "commoditized" "competitive differentiation" "measuring twice if not three times and cutting once." "believe very strongly" "highly committed"

    Tell me a bullshit artist without telling me...anything, as far as I can tell

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like