Trading the stockmarket (NO Referrals)

CNBC's video on this Deepseek development is quite interesting, however it does flag that this is largely piggybacking off the success of GPT 4o. I don't think that this would've been possible without any of the works done by the big names, which is a polite way of saying that China does what China does - nick everyone else's work and then try and profit off it, or advance it in their own way and then claim glory which looks like what is happening here.

I hope I'm wrong and this is in fact a fantastic thing to add to the world of AI but there are too many flags for me, like them not wanting to disclose any info at all about their work.

That being said, I think the whole of silicon valley has just been given an Chinese shaped cricket bat of reality to the face, their exclusive and closed party has just been broken open like a coconut.

For an industry full of questions regarding the openness of their development and morality of what's happening, it's going to be a very interesting few months ahead.
 
Anyone with even the slightest interest in AI and tech can tell you that this AI stuff is just a massive bubble.

These algorithms have existed for decades, they're nothing more than very complicated logic machines operating on an if-this-then-that model.
Its the number of computation per sec that GPU style parallel computing allows that is a game changer. IMO AI will DEFINITELY make grading cancer, reading MRIs/CT scans/Breasts Scans/ Etc, i.e. a whole bunch of diagnostic jobs redundant in a very short period of time. Once you have a system trained up on 1M MRIs or whatever, it's going to make far better determinations than a human, and never have a bad day. FOr example, the determination of low grade dysplasia is very subjective, one person might see it, another might not. It is one of the reasons why a biopsy that comes back with soley LGD is redone.

I would strongly discourage anyone going into the radiographer field.
 
Last edited:
oA1FCAA.png


But the deepseek still uses Nvidia hardware, I can understand AI firms whose tech is AI programs falling but Nvidia shouldn't really be affected given everyone still wants their hardware
 
Broadcom -17%

Damn if i don't laugh I'll cry.

This will probably be my biggest single day of loss "ever"

Oh only about 5%. Pfft
Probably another 5% tomorrow

Yippee
 
Last edited:
In which case, for an idiot like me, how would you explain it?
Neural networks are literally simulations of brains, that learn the same way brains do.

They are definitely not 'if this then that' type algorithms. You don't tell a neural network what to do, you teach it. Which is an incredibly oversimplified summary given the range of models and technologies, but is the crux of it.

Anyone that does a job sat in front of a computer should be thinking how they're going make a living when that job disappears.
 
Neural networks are literally simulations of brains, that learn the same way brains do.

They are definitely not 'if this then that' type algorithms. You don't tell a neural network what to do, you teach it. Which is an incredibly oversimplified summary given the range of models and technologies, but is the crux of it.

Anyone that does a job sat in front of a computer should be thinking how they're going make a living when that job disappears.

Interesting. I'd love to find an actual explanation as to how it works. Might have a looksee on Youtube.
 
In which case, for an idiot like me, how would you explain it?
for object detection they basically learn like a child does from a picture book.

but the picture book has every object inside a rectangle and each rectangle has the name of the item inside.
it's basically just a classifier, I don't know exactly how it learns.

someone's job is to literally label images and videos to AI models can tell what things are.



they don't work like brains though, that's reinforcement learning stuff PPO's etc

they learn the same way as a child.

but the reward/punishment is just a line of code.

like for an agent to play a video game.

taking dmg = -1
damaging an enemy 1.5
standing still too long -0.05

eventually over time they learn the optimal strategy based on the reward system.
tying actions to game states and reward functions etc.

the one that beat a team of humans at DOTA2 had 180 years worth of simulated experience, and probably thousands of reward systems built in.
this videos kinda simple, what the AI sees in a box on the left, it's actions on the right.


chatgpt can mostly come up with the code needed, but takes a lot of trial and error
 
Last edited:
oA1FCAA.png


But the deepseek still uses Nvidia hardware, I can understand AI firms whose tech is AI programs falling but Nvidia shouldn't really be affected given everyone still wants their hardware

Yeah but this is kind of saying
"AI is easy to run on cheap hardware"

So all those GPUs being bought is over supply.
And
Monitisation is going to be really hard for those who've shelled out loads for these gpu farms
And
Nvidia is priced to grow and grow and grow. Any hint of reduction in revenue growth rate and the price will tumble.

Nvidia will be fine. It's current valuation? Less so
 
Yeah but this is kind of saying
"AI is easy to run on cheap hardware"

So all those GPUs being bought is over supply.
And
Monitisation is going to be really hard for those who've shelled out loads for these gpu farms
And
Nvidia is priced to grow and grow and grow. Any hint of reduction in revenue growth rate and the price will tumble.

Nvidia will be fine. It's current valuation? Less so
Time for some Trump legislations to ban all non-American AI use?

Link to government blah blah, link to tiktok etc.
 
Last edited:
For now, they're always going to be chasing performance increases from the hardware to get even more powerful AI that will require the processing units from more powerful hardware to compute

Depends. This system apparently charges so little I'm not sure if the spend on top hardware can ever drive a profit.

All this is so new, but monitisation still seems to be really difficult
 
Hmmmm. From what limited research I've done this evening whilst I'm definitely not supposed to be working, all this AI stuff is still just basic logic. When you boil it down to each "neuron", it's literally just logic being applied and weighted to form overarching decisions. Granted that's a massive oversimplification but it's still not capable of actual reasoning beyond the parameters and predetermined weighting it's fed, and it's not capable of emotion.

Anyway, that's a topic for the AI thread and one I'm definitely going to look more in to.

I posted this in there too but I guess it applies here just as much as it shows how DeepSeek is eating everyone's lunch:


Quite the bloodbath so far, looking forward to seeing where this goes.
 
Depends. This system apparently charges so little I'm not sure if the spend on top hardware can ever drive a profit.

All this is so new, but monitisation still seems to be really difficult
The monetisation comes later and it will most certainly be companies looking to replace humans for AI that can work 24/7 for a fraction of the cost of humans where the only real overhead is rack space and electricity for the hardware to run it all
 
oA1FCAA.png


But the deepseek still uses Nvidia hardware, I can understand AI firms whose tech is AI programs falling but Nvidia shouldn't really be affected given everyone still wants their hardware
It's still needed for training, but not really necessary for running the models. AI in the cloud is prohibitively expense for mass adoption, right now AI is all moving to run on local devices.

I've been running the Deepseek r1 14b model on my PC this evening, happily chugs along just running on the CPU, think it won't quite fit on my 4070, but it works as is.

Edit : Oh no wait, I just had to kill a few things (like Da Vinci Resolve) that were using my GPU. I got the 14b model running on GPU now, all fast like
 
Last edited:
China has been putting stacks of cash into AI for years now, deep seek is the first publicised Chinese AI system… they are likely to be hiding others for government and military use.

Personally I still believe that what we consider AI currently is like the bulletin boards of what the internet used to be.. the UK invented networks and the internet, and America made it to what is is today.. the same thing could happen with AI with another country taking over from Silicon Valley as the flagship leaders.

From what I’ve seen and used of AI at home and work, it’s just frustrating and slows me down.. it’s like having some kid on work experience with you, having to teach them and find them stuff to do. No doubt at some point it will overtake me and I welcome the golden parachute and retirement when the time comes.

Anyone who worked in a modern software engineering environment will understand why AI needs the hardware it does.. back in my college and uni days we was told to program for the lowest hardware possible and to refine our code for performance and reliability. Now a days we just chuck more hardware at it to make it run better, reusing the same source libraries and no one wants to reproduce the wheel, no matter how badly fitted or wonky the wheel is.
If some companies took their time to refine the code…rather than throw more hardware at it, then AI may progress at a fast rate rather than relying on the performance of nvidia hardware, maybe that time will come once gpu development reaches a plateau, but currently nvidia has the best hardware and libraries.

I still think the companies that have the data for AI to process will be more valuable than the companies that process or help with the processing of data in the long run. It’s like the gold rush… yeah the people who sold the mining equipment made shed loads, as nvidia are at the moment.. but the sale of land for mining made more.
 
Now that deepseek has suffered a "massive cyberattack" - will nvidia recover a bit? I made $2 a share on the bounce overnight.
 
Back
Top Bottom