The39thStep
Urban critical thinker
Which seems to be the most important thing about AI
No "The AI bubble has finally burst" thread?
- Thread starter tommers
- Start date
Yeah, why do people need to get so political about stuff?
tommers
It was... Rebekah Vardy's account
It doesn't matter what it actually does. What matters is that they've destroyed the "we need $3 trillion and 500 nuclear reactors to make our slop machines". That shakes everything up.
And don't use GenAI to find out about things. Please. There are very good search engines which can do that already.
And don't use GenAI to find out about things. Please. There are very good search engines which can do that already.
kabbes
First intersubjective, then internalised
There are lots of very different issues being confused and aggregated in this thread. Let me just tease out some separate themes.
1. Is AI actually “useful” and is it going to transform the working world?
Nothing has changed on this score as a result of DeepSeek. DeepSeek does exactly what other LLMs do, just with more claimed efficiency. If you were a believer before, you still will be. If you were not, you still won’t be.
2. Has the AI “bubble” burst?
First, whether or not you think that the whole industry around AI is a “bubble” depends on question 1. And, like I say, nothing has changed regarding question 1. So the answer to this question is kind of ‘no’, because if there’s a bubble then the bubble still exists, and if there isn’t one then there’s no bubble to burst!
Looking in more detail, though, it depends which AI bubble you mean. Let’s disaggregate that…
2a. Has the “bubble” burst in the US tech software sector devoted to creating AI?
This is the sector that includes the likes of OpenAI, and it’s the bit that is now looking most shaky. Basically, there is a big field of players in the US that have spent billions creating their own proprietary LLMs and they’ve just been blindsided by a much cheaper Chinese competitor. A bit like the field of electric car manufacturing, really. They looked like the only game in town and they had valuations to match. Suddenly, they’ve got meaningful competition.
2b. Has the “bubble” burst in the US tech hardware sector devoted to creating AI?
Trickier to answer. This is your likes of Nvidia, the company that makes the computer chips that have been used by the software teams in 2a. They are both deeply affected and also completely unaffected by DeepSeek, depending on what you mean. If DeepSeek can work more efficiently then software teams don’t need to buy so many expensive chips, which is the thing that turbo-charged Nvidia’s growth in 2024. So that’s bad for Nvidia. On the other hand, there’s no reason to think that the software teams won’t end up throwing as much hardware as they can at the problem.
2c. Has the “bubble” burst in the US stock market?
In short, unlikely. If anything, the fact that AI just got cheaper is likely to help the non-tech part of the stock market. If you believe that the answer to question 1 is ‘yes’ then anyone who wants to use AI just got a big boost. That’s why the Dow actually went up yesterday and the NYSE was flat. Today, the European markets have risen.
1. Is AI actually “useful” and is it going to transform the working world?
Nothing has changed on this score as a result of DeepSeek. DeepSeek does exactly what other LLMs do, just with more claimed efficiency. If you were a believer before, you still will be. If you were not, you still won’t be.
2. Has the AI “bubble” burst?
First, whether or not you think that the whole industry around AI is a “bubble” depends on question 1. And, like I say, nothing has changed regarding question 1. So the answer to this question is kind of ‘no’, because if there’s a bubble then the bubble still exists, and if there isn’t one then there’s no bubble to burst!
Looking in more detail, though, it depends which AI bubble you mean. Let’s disaggregate that…
2a. Has the “bubble” burst in the US tech software sector devoted to creating AI?
This is the sector that includes the likes of OpenAI, and it’s the bit that is now looking most shaky. Basically, there is a big field of players in the US that have spent billions creating their own proprietary LLMs and they’ve just been blindsided by a much cheaper Chinese competitor. A bit like the field of electric car manufacturing, really. They looked like the only game in town and they had valuations to match. Suddenly, they’ve got meaningful competition.
2b. Has the “bubble” burst in the US tech hardware sector devoted to creating AI?
Trickier to answer. This is your likes of Nvidia, the company that makes the computer chips that have been used by the software teams in 2a. They are both deeply affected and also completely unaffected by DeepSeek, depending on what you mean. If DeepSeek can work more efficiently then software teams don’t need to buy so many expensive chips, which is the thing that turbo-charged Nvidia’s growth in 2024. So that’s bad for Nvidia. On the other hand, there’s no reason to think that the software teams won’t end up throwing as much hardware as they can at the problem.
2c. Has the “bubble” burst in the US stock market?
In short, unlikely. If anything, the fact that AI just got cheaper is likely to help the non-tech part of the stock market. If you believe that the answer to question 1 is ‘yes’ then anyone who wants to use AI just got a big boost. That’s why the Dow actually went up yesterday and the NYSE was flat. Today, the European markets have risen.
The39thStep
Urban critical thinker
Didn't most people who aren't that political know that there is censorship in China before there were screenshots of political people typing Tianaman Square into DeepSeek to prove there is censorship in China ?
ska invita
back on the other side
yeah this development is boring now - has it even taken down any hedge funds? seems not
BristolEcho
Well-Known Member
The AI sector in America has been straining for a while hasn't it? Hence the constant "We are making lots of progress" claims. So for them this is not good news.
This is me being totally clueless on stocks but I have a question. Nvidia were already a massive company before the recent spike due to AI. Obviously losing so much in one day isn't good, but they are still valued much higher than they were a year ago if I'm following correctly. This is bad for the people holding stock hoping it would rise, but for the company itself what does this mean? It's bad as they won't be able to sell as many chips, and I guess they will have less power to buy/leverage money for other projects? They are still going to be making vast amounts of profit I imagine? I guess it's the mentality of "line must go up" but I'm sure the people involved in the company knew that wouldn't last forever.
burtabraham
Well-Known Member
It makes it even more likely that AI will transform the UK economy (and society, culture, and values), and it was already 99% certain. It just means that AI is a lot cheaper and more efficient than we thought.
My prediction is we will see many more events like this in the next decade. We really are at the beginning of a new Industrial Revolution. Nothing will be the same again, so buckle up!
kabbes
First intersubjective, then internalised
Right. And they are a very specific niche of the wider marketplace, and what is bad for them might be good for others.
A full treatment of this subject would take a lot of room. But there are effectively two key reasons why a 'company' cares about its share price, bearing in mind that by 'company', we really mean 'the management of that company, whose interests may or may not align with other stakeholders'.
First, companies sometimes want to raise more money so that they can do things with it, like build new factories. The higher their share price, the easier and cheaper it is for them to raise these new funds.
Second, financial capitalism is EXTREMELY predatory. If your share price drops too much, you become vulnerable to a hostile take-over. Regardless of the goods and bads of this, it is unlikely that you want it to happen if you are the senior management.
The Guardian looked into DeepSeek censorship and found some workarounds
View attachment 461598
We tried out DeepSeek. It worked well, until we asked it about Tiananmen Square and Taiwan
The AI app soared up the Apple charts and rocked US stocks, but the Chinese chatbot was reluctant to discuss sensitive questions about China and its governmentwww.theguardian.com
I'm glad to see bloated US tech giants get a bloody nose but not sure if a cheap alternative from a country controlled by a totalitarian regime is a great development
I need to learn more about how these things are trained, from a layman's PoV. But as it's open source, could someone else just not edit and compile a version that is not so sensorious?
*. Actually to anyone who knows, can in fact anyone run this and train it on their own material?
Is that what we're looking at?
*. I tried GPT4All for this sort of thing but the interface is pretty inaccessible to a screen reader, so never got it to do anything. Still have it installed on this machine. This could be actually really useful if this is how it works and you don't need 100GPUs and a TB of ram.
Last edited:
Fez909
toilet expert
It doesn't affect Nvidia at all, really. They have so much cash, that they don't need to raise funds. They actually have problems spending the money they have. Apple are in a similar position.
It has an impact on the employees and directors who own the shares. Lots of developers are paid in stock options and many of them became insanely wealthy when their shares went up. The CEO has been selling his shares in a steady stream for the past few month. Not sure how many, but he's sold hundreds of millions of dollars worth. Of course, if the stock price goes down, he gets less for them. But it's still up massively from last year, so this is not really an issue for him or other insiders.
The people who will be most affected as people who bought shares recently - obviously their investments are currently losing money. And this includes large funds, including pension funds (I think they're called 501Ks in USA). So if you cashed in your pension yesterday, you would have a smaller pension pot.
BristolEcho
Well-Known Member
Thanks both!
kabbes
First intersubjective, then internalised
Fez's answer made me realise that I missed another important reason why a 'company' cares about its share price. Attraction and retention of staff -- if you are incentivising people to stay with you based on the golden handcuffs of unvested shares then it's important that those unvested shares are worth a lot of money.
newme
Giant in Pastyland now
Apple is holding lots of money out of the US iirc. Subject to taxes if brought over so of course they don't. Was a while back I heard this tho. Is this similar for nvidia? Know TSMC are making some fabrication facilities in the US now. Nvidia being fabless I assume contributes.
PR1Berske
Alligator in chains by the park gates.
And this is why the bubble will burst. Advocates will have to try and justify censorship and political biases, or wave them away, neither of which seem good for the long term.
If you want to ask about Hong Kong and the "best" resource can't tell you because of government decree, then it's not a resource. Put AI into schools, as advocated by a minority at the moment, and you're giving them misleading or politically insensitive misinformation.
It will be interesting to see how it all pans out. Google's been banned in China for years because of its refusal to follow the Party's decrees on what's acceptable to search for. But that bubble's not exactly burst.
lazythursday
Averagely-Known Member
It's not political bias that is the issue in my opinion. It will be the unintended biases that emerge from using AI to process data at scale, eg in the benefits system, or life insurance etc, hat will undoubtedly harm groups of people. And the related fact that it consistently hallucinates and gets things wrong and that will also harm outcomes. You can argue that humans do too - and perhaps a 'mostly reliable' AI is as good as a human, but will need a whole load of safeguards to prevent risk and harm to people (and as an extension, legal and financial risk to individuals companies and institutions).
kabbes
First intersubjective, then internalised
And the fact that many of these problems are explicitly prohibited under GDPR. There’s a clash of governance on the horizon right there.
two sheds
Least noticed poster 2007
previously posted to the wrong thread 
www.theguardian.com
Experts urge caution over use of Chinese AI DeepSeek
UK officials say they are monitoring any national security threat to data from the new AI
Slightly aware as I am of Greek & Trojan history, I'm not sure I'd trust downloading the program onto my computer for running 'offline' for similar reasons.
Saul Goodman
It's all good, man
You have to assume that every piece of software you install is spying on you. If it's from China, you have to assume it's spying on you for the Chinese government. Especially so if it's free.previously posted to the wrong thread
Experts urge caution over use of Chinese AI DeepSeek
UK officials say they are monitoring any national security threat to data from the new AI
Slightly aware as I am of Greek & Trojan history, I'm not sure I'd trust downloading the program onto my computer for running 'offline' for similar reasons.
Saul Goodman
It's all good, man
I doubt they'll do that. They won't need to when everyone willingly gives it any information it asks for
The39thStep
Urban critical thinker
You never know with the 'yellow peril '.
newme
Giant in Pastyland now
On open source? If there is it will be found.
But yeh pretty much assume any time you do anything at all someone is stealing data somehow, makes it far easier when thats the case. Also far more careful otherwise as a result, tho it's almost all irrelevant anyway cos its anonymised as a data set the majority of the time. Then when it isn't the bank refunds you and was probably someone else actively trying to do this rather than it being anything you did. Except if you bought something off some website that you never heard of that was surprisingly cheap, or similar offline.
two sheds
Least noticed poster 2007
UnderOpenSky
baseline neural therapy
Sorry, poor answer from my phone whilst cooking dinner. My understanding is it's impossible for these things to be open source in the traditional meaning, the models can be open source, but it's the training data that makes the magic and that isn't. I'm happy to stand corrected, but we still don't totally know how these things train themselves. So you can start with two open source models and end up with very different results.
And back to your original point, even if it starts open source, we don't know what's happening with that data.
Edit. These people explain it far better then I ever could
2.5 Admins 211: Open Sourceless – 2.5 Admins
2.5admins.com
Crispy
The following psytrance is baṉned: All
I take it back, the whole thing is about 800GB and you need to load it all into RAM. So it's not for the home user at all.
Dom Traynor
Tino Pai
gosub
~#
A-Lie then
Similar threads
