Won't somebody think of the CEOs?
Sam Altman rolling around on his mattress stuffed with specially printed million-dollar bills, the only thing he can sleep on, screaming like a toddler who has just been given the wrong kind of biscuit.
Share prices for some of the biggest American tech brands that crested the AI hype waves crashed this morning on the rocks of DeepSeek, a Chinese startup that last week released LLMs that challenges US dominance. As The Register revealed at the weekend, DeepSeek launched some openly available machine-learning models which …
It can't pop soon enough.
‘Serious concerns’ about DWP’s use of AI to read correspondence from benefit claimants
The DWP apparently thinks it can shovel everything into the magic AI machine and it'll do everything for them:
That is the challenge facing the Department for Work and Pensions (DWP) as correspondence floods in from benefit applicants and claimants – of which there are more than 20 million, including pensioners, in the UK. The DWP thinks it may have found a solution in using artificial intelligence to read it all first – including handwritten missives.
Human reading used to take weeks and could leave the most vulnerable people waiting for too long for help. But “white mail” is an AI that can do the same work in a day and supposedly prioritise the most vulnerable cases for officials to get to first.
By implication, it deprioritises other people, so its accuracy and how it reaches its judgments count, but both matters remain opaque. Despite a ministerial mandate, it is one of numerous public sector algorithms yet to be logged on the transparency register for central government AIs.
[..]
An internal data protection impact assessment said letter writers “do not need to know about their involvement in the initiative”.
The assessment says correspondence can include national insurance numbers, dates of birth, addresses, telephone details, email addresses, details of benefit claims, health information, bank account details, racial and sexual characteristics, and details on children such as their dates of birth and any special needs.
People who work with benefit claimants are now voicing “serious concerns” about how the system handles sensitive personal data.
As I said, there is no business case for spending zillions on this.
Trump might ban the Chinese version or stick 10,000% tariffs on it, but there is another way and all those promised vast investments just vanished.
If you buried a tonne of cash in AI, let's hope you learned from the Metaverse and the small print included a few lines about clawing it back if you changed your mind.
10000% tariffs wouldn't be a problem - it's free to download and MIT licensed. I suspect it's been downloaded at least a million times by now, and people will spend a week or so struggling with getting Docker to work and installing and running it, then another week playing with it until they get bored. Then it will be available on umpteen websites for the public to do the same with, and then even more people will conclude that it's been overhyped, which will cause share prices to fall even further.
Any business case that might have existed before 20th Jan 2025 MUST be rewritten if not tossed into the garbage. Since the advent of Trump 2.0, his memecoin (and that of Melania) has tanked. The world economy is preparing for his tariff wars and the inevitable recession that might lead to depression.
With a deep recession ALL spending will be cut back including AI.
Perhaps it it time to cancel those exotic holidays?
Banning China from using US technology will only ensure China develops their own and will ban the US & al. from using it.
They do really smart stuff in East and South Asia. The fact that much of the smart work in US universities and US tech is done by people who were educated in East Asia should have been a warning sign.
Wishful ThinkingTM is not a good basis for policy.
Didn't we have exactly this Tech race with encryption keys? The US Government would not allow any US companies from selling keys over 64(?) bit length to non-US customers so some Johnny Foreigner wrote their own 128(?) bit version and threatened to not sell it to US costomers
Microsoft, for example, dumping in tens of billions and then lying out their ass about how much money they're making on it by counting anyone who has Copilot or Recall installed as 'AI revenue'.
The only thing that will possibly make back the money they poured into the 'AI' toilet is AGI, and good luck with that, you won't get it with LLMs. And if you suddenly had AGI, now you've got a whole new set of existential crises, but as usual they think they'll be the ones fully in control, what could go wrong?
I suspect the thinking about AI is that AI will replace human employees. In this case, the supplier who controls the AI can cream off the profits from firing those employees.
That has been MS' business model with Windows & Office. Quite natural that they expect to repeat this with AI.
But this only works when you are a monopolist. The US has all the, IP based, legal Infrastructure to enforce a US monopoly. The tariff insanity has been installed because the "foreigners" are starting to win the IP game in technology.
Becoming a global monopolist in AI has become much harder with the arrival of a credible non-American competitor.
Indeed, Pop! is likely.
>That has been MS' business model with Windows & Office. Quite natural that they expect to repeat this with AI.
Ironically MS business model has been that companies need 1000s of office workers all working in Office and all paying $/month for subscriptions.
If MS really believed that AI was going to replace all the workers at their customers - they would go totally "Balmer" on AI
"If MS really believed that AI was going to replace all the workers at their customers - they would go totally "Balmer" on AI"
Controlling AI also means controlling the pace it will replace Office.
Remember how IE was used to stop progress and innovation on the web and internet. Only after IE's strangleholds on the browser market was broken could the cloud revolution start.
"There were so many breakthroughs that lead to the cloud revolution"
Indeed, but the first step was easy access to services for the masses, which required a capable browser. And IE was many things, but not capable. First and foremost, IE had its wings clipped to prevent any access to other companies' services.
Only after Google forced Chrome into the market were they, or anyone else, able to develop real cloud based web services.
MS business model is stepped change with big associated costs payable to MS each time. So...
- Everyone needs a PC, no big mainframes
- Every PC needs to be on a network with a server
- No now do it in the cloud (looks a lot like stage 1)
- TPM 1.0 and now you need Win 10 to support it, conveniently we won't support previous versions any more.
- TPM 2.0 and now you need Win 11 now we've got you locked in to TPM.
- Now you all need AI
There's good arguments (the best that MS can pay people to come up with) for each step, but they all funnel cash to MS (which is the real argument).
As for profit, you could start by charging based on actual costs. Within 20 % of break even, say, for usage where the data is not also being used for training.
Especially in the gratuitous pleasure market, e.g., home generation of graphics, music, unmentionable, there is really no justification to be taking a loss on those, yet OpenAI has set low prices for those uses simply because that's is all the market would bear. Never mind deep seek, the writing is right there on the wall already.
Resource constraints and competition drives innovation. If Deep Seek is really 10x better, that is how they got there. The current US model of a king and his handful of tech lords has become the anti thesis of capitalism - it harks back to the medieval economy. I am exaggerating, but that trend is real.
I would argue that the problem is not AI per se, but the pseudo capitalist ego baring way in which it is developing.
If you have too much resource then what tends to happen is you waste by throwing lots at things without thinking whether the results will be worth it, when your resources are more constrained then you think more carefully to avoid squandering them and are more likely to look for inventive approaches.
The big players in AI seem to have been brute forcing it and that often has exponentially decreasing results for the extra effort.
What is needed is enough resources that people feel doable and can take a few gambles, but not so much that they will end up wasting huge amounts just trying things.
Is there any point now in the export controls the US have been putting on the GPUs that could be used by China to train AI, if they have already shown what they have done has beaten the US at their own game?
Because if Deepseek have managed to do this for $5m with the GPUs they were allowed, then im damn sure the Chinese government will have already got access to this tech for military uses by now.
But if they've managed to get access to Nvidia's latest GPUs via front purchases maybe by multiple shell companies they set up elsewhere (some maybe even in the US) they would hardly want to admit that, or risk having their supply cut off.
Claiming they accomplished it with much older/cheaper tech avoids that and has the added benefit of hitting high flying US tech companies which would count as a double win in Xi's book.
I don't know which is true, but I don't know how you could possibly keep that sort of thing from happening. It isn't as if Nvidia goes and checks on the companies that want to buy their products to insure they truly exist in a form that can use them. If the wire transfer completes they're going to be happy to deliver the product, and not do a whole lot of investigation. Heck I don't know how you could tell the difference between a shell company front and an AI startup. Neither will have any corporate history, any products, maybe not even a real office as I'm sure some startups these days are 100% remote and have their "corporate headquarters" listed as being in some attorney's office.
Nvidia aren't selling millions of H100 gpus to randoms on ebay. Even Twitter could only get them by having Musk steal them from Tesla.
They may have got a handful to play with through 'unofficial channels' but not the 100,000 GPU clusters that OpenAI and chums are talking about needing.
They didn't get "access to NVIDIA's latest GPUs."
They got old ones, H800, dedicated to the China market. [1]
Only 2048. Putting all those together into a real AI-capable machine would cost around $400M, including the H800s. (Using prices from 2 years ago, which is when they would have built such a machine.)
Now, $400M is a lot, but it seems to be a lot less than what OpenAI spent on their infra, and whatever Elon is spending.
The Deepseek model build was cheaper, but not cheap.
[1] https://www.theregister.com/2024/04/23/china_banned_nvidia_gpus/
Deepseek’s parent is a quant fund, apparently. The impact of their reveal has been to carve $500bn off of Nvidia’s stock price.Has anyone checked to see if Deepseek’s parent held short positions in Nvidia? .... skwdenyer
In the words of the inimitable Bart Simpson .... Eat my shorts! ..... And who would have a’thunk it, if true, .... a Middle Kingdom quant fund leading and raiding and repurposing global markets the tried and trusted/tested and celebrated crazy western capitalist market way.
Bravo, Encore, encore. What’s not to like in such a disruptive change and novel situation..... for is not the market rigged to ensure too big to fail doesn’t fail even whenever its product is destined to not succeed and support innovative future growth and increasingly rewarding shared prosperity?