Ed Zitron has been writing about the many structural problems with the Silicon Valley AI/LLM business model for well over a year now.
DeepSeek has punctured the OpenAI/Anthropic side, now we wait to see how ASICs or custom code H800s or whatever punctures the hardware side.
But the core question still remains for LLMs: are they really all that useful?
I have repeatedly noted that LLMs are different than all of computing to this point. For all of computing history up until now, you only got garbage out, if you put garbage in (GIGO).
LLMs can spout garbage at any time for any reason.
And when this dynamic exists, then it is necessary to constantly check - at what percentage of GO (garbage out) do the supposed benefits of LLMs go away?
I have no doubt money is going to be made on LLMs - after all, look at crypto. But profit made does not equate to productivity increases or economic benefits to society as a whole.
5 million? Really? Are you sure it wasn't $500 million of smuggled chips? I know the CCP would never lie about anything, but maybe just this once they told a little fib 🤔
Deepseeks own paper says they basically just built a copy of other AI models (although not the formal announcements), none of which is anything fancy or innovative, most western expertise knows about the approach, just normally prevent from doing it by "terms of use". Potentially that may gain might be rewriting the base code that runs the GPU, they do have a former Nvidia engineer working for them, so this would be possible. I know some programmers on X been trying to do the same thing for a while but don't have much support from Nvidia to do this, but this isn't really an innovation, this is just bypassing a gatekeeper using inside knowledge. Though it appears to know be challenging major companies to start looking at doing the same. Finally all this ignores all the cybersecurity, malware and other issues that have been identified in the last few days (not that CHatGPT was super great at this either but they were better. In the end I don't see the gains from it other effective releasing the ChatGPT model for free rather than having to pay for it (tons of gains for the hedge fund backing it though who had a ton of shorts on NVidia prior to the announcement lol). It just has surprised me how few people seem to have done much research into it and have accepted the release story at face value, particularly when there are a ton of that were at least obvious to me after I read the released paper and found about the short position. Nothing against seeing OpenAI sweat a little but also need to understand what is actually happening. Deepseek could never have done what they did without OpenAI already having a model, so it obviously requires all OpenAI learning to get to where it got too, so basically they didn't do anything cheaper, just showed everyone how to cheaply copy someone elses model which will probably cause others to lock down there own models more in the end.
The reasons why DeepSeek has been cost efficient in the development of the superior AI platform, are very simple.
China does not possess the luxury of reserve currency status, with the largest equities markets in the world fueled by comical amounts of currency creation from thin air. Then leverage that phony currency 100:1 in loans to banks, which then loaned it to businesses at near ZERO % interest, and you have massive distortions in investment risk management.
When 'money' is free, blowing $250 Billion to invest in marginally beneficial technology which it's purpose is merely to catalogue existing information, compile it, and regurgitate it back to people whilst sounding brilliant, just shows the value of that 'money' is essentially worthless.
Unless of course you're use of AI is creating female dream queens that you put on OnlyFans and charge thirsty men $10 a month to look at your fake sexiness, then in this case AI is brilliant.
When you are sanctioned, maligned and impugned constantly as well as your currency not widely traded and certainly not in reserve status, you tend to be scrappy and look for maximum bang for the buck so to speak. That apparently is DeepSeek and China as a whole if you're being honest with yourself.
Will be interesting to see how the market continues to perform its reality check.
Ed Zitron has been writing about the many structural problems with the Silicon Valley AI/LLM business model for well over a year now.
DeepSeek has punctured the OpenAI/Anthropic side, now we wait to see how ASICs or custom code H800s or whatever punctures the hardware side.
But the core question still remains for LLMs: are they really all that useful?
I have repeatedly noted that LLMs are different than all of computing to this point. For all of computing history up until now, you only got garbage out, if you put garbage in (GIGO).
LLMs can spout garbage at any time for any reason.
And when this dynamic exists, then it is necessary to constantly check - at what percentage of GO (garbage out) do the supposed benefits of LLMs go away?
I have no doubt money is going to be made on LLMs - after all, look at crypto. But profit made does not equate to productivity increases or economic benefits to society as a whole.
And next up, we might just find AI is mostly hype with no real benefit.
And China just revealed that money is the ultimate illusion of value; again.
5 million? Really? Are you sure it wasn't $500 million of smuggled chips? I know the CCP would never lie about anything, but maybe just this once they told a little fib 🤔
Deepseeks own paper says they basically just built a copy of other AI models (although not the formal announcements), none of which is anything fancy or innovative, most western expertise knows about the approach, just normally prevent from doing it by "terms of use". Potentially that may gain might be rewriting the base code that runs the GPU, they do have a former Nvidia engineer working for them, so this would be possible. I know some programmers on X been trying to do the same thing for a while but don't have much support from Nvidia to do this, but this isn't really an innovation, this is just bypassing a gatekeeper using inside knowledge. Though it appears to know be challenging major companies to start looking at doing the same. Finally all this ignores all the cybersecurity, malware and other issues that have been identified in the last few days (not that CHatGPT was super great at this either but they were better. In the end I don't see the gains from it other effective releasing the ChatGPT model for free rather than having to pay for it (tons of gains for the hedge fund backing it though who had a ton of shorts on NVidia prior to the announcement lol). It just has surprised me how few people seem to have done much research into it and have accepted the release story at face value, particularly when there are a ton of that were at least obvious to me after I read the released paper and found about the short position. Nothing against seeing OpenAI sweat a little but also need to understand what is actually happening. Deepseek could never have done what they did without OpenAI already having a model, so it obviously requires all OpenAI learning to get to where it got too, so basically they didn't do anything cheaper, just showed everyone how to cheaply copy someone elses model which will probably cause others to lock down there own models more in the end.
The reasons why DeepSeek has been cost efficient in the development of the superior AI platform, are very simple.
China does not possess the luxury of reserve currency status, with the largest equities markets in the world fueled by comical amounts of currency creation from thin air. Then leverage that phony currency 100:1 in loans to banks, which then loaned it to businesses at near ZERO % interest, and you have massive distortions in investment risk management.
When 'money' is free, blowing $250 Billion to invest in marginally beneficial technology which it's purpose is merely to catalogue existing information, compile it, and regurgitate it back to people whilst sounding brilliant, just shows the value of that 'money' is essentially worthless.
Unless of course you're use of AI is creating female dream queens that you put on OnlyFans and charge thirsty men $10 a month to look at your fake sexiness, then in this case AI is brilliant.
When you are sanctioned, maligned and impugned constantly as well as your currency not widely traded and certainly not in reserve status, you tend to be scrappy and look for maximum bang for the buck so to speak. That apparently is DeepSeek and China as a whole if you're being honest with yourself.
Will be interesting to see how the market continues to perform its reality check.
It make's me happy, kicking all those trillionaires up the proverbial butt!
its not open source
As scary as watching robots dance!
For a long time, many of us knew that computing power was the real war. Now we know for sure!
Spot on, Rich!