No "The AI bubble has finally burst" thread?

[tommers]

they would have presumably copied open source code though?

Well that article goes into the details but basically you can use existing models to train new ones.

Not ideal for obvious reasons but cheap.

Which is probably the real shock of this, it's cheap and it uses cheaper GPUs, less power etc etc. Whereas before Nvidia were like "you need to just send us money and everything will be fine".

 

[JimW]

Reminds me of the stories of the PLAAF shooting down the U2 spy planes back in the day, low tech cleverness and effort a match for expensive kit.

 

[The39thStep]

On Deepseek I got this

 

[PR1Berske]

Is that going to be .... errr .... secure* if it's a Chinese firm?

* not sure what that even means with AI, or with the alternative being American firms

It's been labelled "the Sputnik moment" and with good reason. Potentially "better" product despite (in spite?) of US sanctions on exporting chips to China.

The bubble point is still open for debate. AI is a fad, absolutely it is, and there's no valid reason for any company to use AI. Not one valid reason. It's the NFT du jour. If China has created AI which actually does something important is an open question for me.

 

[Saul Goodman]

It refused to converse with me



ChatGPT had no such compunction

 

[two sheds]

I just learned something I'd rather not have known

 

[gosub]

It's a FAD that's driven the stock exchanges for over year ...if you took the big 6 out of the Dow was quite lean times

 

[Yossarian]

Wonder if Trump is going to try to ban DeepSeek, would probably be easy enough to do so with the TikTok precedent - the AI's probably got a "flatter Trump" mode that can be activated if a ban appears likely

 

[pseudonarcissus]

AI's time is up...

I’d say no-quite the opposite. AI is becoming more integrated into everyday life and various industries, from healthcare to entertainment. While there are ongoing debates about regulation, ethics, and risks, advancements continue at a rapid pace

 

[JimW]

All the content restrictions are immaterial if someone else takes the open source code and runs it independently, surely?

 

[Pickman's model]

Wonder if Trump is going to try to ban DeepSeek, would probably be easy enough to do so with the TikTok precedent - the AI's probably got a "flatter Trump" mode that can be activated if a ban appears likely

And a flatten trump mode if it is imposed

 

[ska invita]

The energy use aspect is also great news

All the content restrictions are immaterial if someone else takes the open source code and runs it independently, surely?

you still need to provide your own trianing data though supposedly

hard for sure to know whats what but I have full trust in the market valuations

 

[JimW]

The energy use aspect is also great news

you still need to provide your own trianing data though supposedly

hard for sure to know whats what but I have full trust in the market valuations

Just a bit of retraining work?

 

[Saul Goodman]

Just a bit of retraining work?

And a few spare data centres, and a hydroelectric power station.

 

[JimW]

And a few spare data centres, and a hydroelectric power station.

Could stick one by the fish pond, no problem.

 

[ska invita]

Just a bit of retraining work?

yes and I think Im right in saying theres things like LEGAL AI (specialist AI, I forget the correct term of that now) where you give the AI every legal document and it really knows that stuff and that stuff only....so for building specific clients then its this kind of open source code that becomes preferable over a more expensive data hungry made in the USA one.

 

[tommers]

yes and I think Im right in saying theres things like LEGAL AI (specialist AI, I forget the correct term of that now) where you give the AI every legal document and it really knows that stuff and that stuff only....so for building specific clients then its this kind of open source code that becomes preferable over a more expensive data hungry made in the USA one.

one of the major innovations is that they've split the model into "experts" and only use the ones needed to answer the query. Rather than trying to use the whole thing for everything.

 

[tommers]

you can run it offline btw.

 

[JimW]

you can run it offline btw.

Stick it on a USB?

 

[Spymaster]

AI is a fad, absolutely it is and there's no valid reason for any company to use AI.

Whaaaaaaat?

It most definitely isn't. It's incredibly transformative tech that's already making a massive impact in loads of industries.

Self driving cars, drones, recommendation engines, chatbots, marketing analytics, robotics ... there are hundreds of reasons for companies to use it. Whether or not that's a good thing is questionable but the notion that it's a fad is nuts. The content creation industry is being hammered by AI right now.

 

[Crispy]

The energy use aspect is also great news

you still need to provide your own trianing data though supposedly

No, training is something only the big boys can do. It takes a long time and costs millions in GPU time.
What this does is makes it much cheaper to do the end-user bit, the "fancy autocomplete" bit.

 

[two sheds]

you can run it offline btw.

what data does it then use?

 

[ska invita]

No, training is something only the big boys can do. It takes a long time and costs millions in GPU time.

you cant trust youtubers

 

[tommers]

what data does it then use?

 

[Saul Goodman]

Whaaaaaaat?

I'm pretty sure he's trolling

 

[Crispy]

what data does it then use?

It's already "used" data in training (and it'll be the same data as everyone else ie. the entire internet). What they've released is the end result - the billions of weights for the nodes in the network. That is static and unchanging for every LLM, simply because deriving those numbers is so astronomically expensive and time-consuming.

PS: This is why LLMs are a complete dead end for AGI. It's like taking a backup copy of a person, telling them a story and ask them to think of the next word, which you write down. Then you kill them and reboot a new version of that old backup and tell them the slightly longer story, get the next word, kill, rinse repeat. The person is unchanging and only every knows their upbringing and the story you just told them. They don't learn.

 

[The39thStep]

I asked the Hong Kong 2019 question and what it does is give an quite lengthy appraisal and summary very quickly but then that disappears as soon as it's finished and I get the ' sorry I can't help message'. I've done this twice ,can't post a video up as the app is on this new tablet I've got and I've only just learnt screenshots.

On Tianaman Square it originally says can't help but then if you say tell me something what you know it gives a short answer about what it is named after and where it is .

So for those who want Deepseek for information about those aspects you'll be better off on another AI , Google or a browser .

 

[The39thStep]

I don't know if this has been posted up elsewhere on this site however Micheal Roberts the economist did a two part write up on his blog of the Allied Social Science Associations (ASSA) meeting (held each year by the American Economic Association ) , this years subject was AI .Covers a wide range of views on a number of aspects including impacts on employment etc.

ASSA 2025: part one – AI, AI, AI..

Every year, I report on the annual proceedings of ASSA, the Alliance of Social Science Associations, run by the American Economics Association. It is the biggest economics conference in the world, …

thenextrecession.wordpress.com

 

[Yossarian]

So for those who want Deepseek for information about those aspects you'll be better off on another AI , Google or a browser .

No shit

 

[The39thStep]

No shit

Thought it might save you from posting up screenshots from social media tbh.