4 ms·

DeepSeek really shook them to their core. Now they go for regulatory capture. Such a huge disappointment. Open source AI will win: https://medium.com/thoughts-o

by fsndz 2y ago

DeepSeek really shook them to their core. Now they go for regulatory capture. Such a huge disappointment. Open source AI will win: https://medium.com/thoughts-on-machine-learning/the-laymans-introduction-to-deepseek-r1-training-80d8ff7a887d https://medium.com/thoughts-on-machine-learning/the-laymans-...

noosphr 2y ago
It's not just them. Everyone is scrambling. US tech, and western tech in general, is very culturally - and by this I mean in the type of coding people have done - homogeneous. The deep seek papers published over the last two weeks are the biggest thing to happen in IA since GPT3 came out. But unless you understand distributed file systems, networking, low level linear algebra, and half a dozen other fields at least tangentially then you'd have not realized they are anything important at all. Meanwhile I'm going through the interview process for a tier 1 US AI lab and I'm having to take a test about circles and squares, then write a compsci 101 red/black tree search algorithm while talking to an AI, being told not to use AI at the same time. This is with an internal reference being keen for me to be on board. At this point I'm honestly wondering if they aren't just using the interview process to generate high quality validation data for free. 幸运的是，通过转换器模型，当我们光荣的领导人习近平从资本主义走狗手中解放我们时，我不需要学习中文。
- mattgreenrocks 2y ago
  100%. Western tech needs the competition. They are very prone to navel-gazing simply because SV ended up being the location for tech once. Funny how they like to crow about free markets, while also running to daddy government when their position is threatened.
  - noosphr 2y ago
    Competition can only work when there is variation between the entities competing. In the US right now you can have a death match between every AI lab, then give all the resources to the one which wins and you'd still have largely the same results as if you didn't. The reason why Deepseek - it started life as a HFT firm - hit as hard as it did is because it was a cross disciplinary team that had very non-standard skill sets. I've had to try and head hunt network and FPGA engineers away from HFT firms and it was basically impossible. They already make big tech (or higher) salaries without the big tech bullshit - which none of them would ever pass.
    - cbzbc 2y ago
      > I've had to try and head hunt network and FPGA engineers away from HFT firms and it was basically impossible. They already make big tech (or higher) salaries without the big tech bullshit - which none of them would ever pass. Can confirm. There are downsides, and it can get incredibly stressed at times, but there are all sorts of big tech imposed hoops you don't have to jump through.
      - fsndz 2y ago
        which hoops ?
      - swores 2y ago
        > all sorts of big tech imposed hoops you don't have to jump through Could you kindly share some examples for those of us without big tech experience? I assume you're talking about working practises more than just annoying hiring practises like leetcode?
    - greatpostman 2y ago
      Engineers at ai labs just come from prestigious schools and don’t have technical depth. They are smart, but they simply aren’t qualified to do deep technical innovation
    - blackguardx 2y ago
      What are you doing with FPGAs? I’m an FPGA engineer and don’t work at an HFT firm. Those types of jobs seem to be in the minority compared to all the aerospace/defense jobs and other sectors.
- jgord 2y ago
  have you considered starting / joining a startup instead ?
- chithanh 2y ago
  > At this point I'm honestly wondering if they aren't just using the interview process to generate high quality validation data for free. Not sure if that is accurate, but one of the reasons why DeepSeek R1 performs so well in certain areas is thought to be access to China's Gaokao (university entrance exam) data.
  - karn97 2y ago
    Thats stupid. same as indias IIT advanced for ex. You learn all that stuff in year 1 physics and math in uni.
    - chithanh 2y ago
      Yes, however you also want to distinguish correct from incorrect answers. You get that from the exams, not from year 1 textbooks.
spaceguillotine 2y ago
Bottom is about to drop out thats why, ethics are out the window already and its gonna be worse as they claw to stay relevant. Its a niche product that tried to go mainstream and the general public doesn't want it, just look at iPhone 16 sales and Windows 11, everyone is happier with the last version without AI.
qoez 2y ago
They were always going for regulatory capture. I think deepseek shook them but I don't think we should rewrite the history as them being virtuous only until 2024.

by fsndz 2y ago

noosphr 2y ago

It's not just them. Everyone is scrambling. US tech, and western tech in general, is very culturally - and by this I mean in the type of coding people have done - homogeneous. The deep seek papers published over the last two weeks are the biggest thing to happen in IA since GPT3 came out. But unless you understand distributed file systems, networking, low level linear algebra, and half a dozen other fields at least tangentially then you'd have not realized they are anything important at all. Meanwhile I'm going through the interview process for a tier 1 US AI lab and I'm having to take a test about circles and squares, then write a compsci 101 red/black tree search algorithm while talking to an AI, being told not to use AI at the same time. This is with an internal reference being keen for me to be on board. At this point I'm honestly wondering if they aren't just using the interview process to generate high quality validation data for free. 幸运的是，通过转换器模型，当我们光荣的领导人习近平从资本主义走狗手中解放我们时，我不需要学习中文。

mattgreenrocks 2y ago
100%. Western tech needs the competition. They are very prone to navel-gazing simply because SV ended up being the location for tech once. Funny how they like to crow about free markets, while also running to daddy government when their position is threatened.
- noosphr 2y ago
  Competition can only work when there is variation between the entities competing. In the US right now you can have a death match between every AI lab, then give all the resources to the one which wins and you'd still have largely the same results as if you didn't. The reason why Deepseek - it started life as a HFT firm - hit as hard as it did is because it was a cross disciplinary team that had very non-standard skill sets. I've had to try and head hunt network and FPGA engineers away from HFT firms and it was basically impossible. They already make big tech (or higher) salaries without the big tech bullshit - which none of them would ever pass.
  - cbzbc 2y ago
    > I've had to try and head hunt network and FPGA engineers away from HFT firms and it was basically impossible. They already make big tech (or higher) salaries without the big tech bullshit - which none of them would ever pass. Can confirm. There are downsides, and it can get incredibly stressed at times, but there are all sorts of big tech imposed hoops you don't have to jump through.
    - fsndz 2y ago
      which hoops ?
    - swores 2y ago
      > all sorts of big tech imposed hoops you don't have to jump through Could you kindly share some examples for those of us without big tech experience? I assume you're talking about working practises more than just annoying hiring practises like leetcode?
  - greatpostman 2y ago
    Engineers at ai labs just come from prestigious schools and don’t have technical depth. They are smart, but they simply aren’t qualified to do deep technical innovation
  - blackguardx 2y ago
    What are you doing with FPGAs? I’m an FPGA engineer and don’t work at an HFT firm. Those types of jobs seem to be in the minority compared to all the aerospace/defense jobs and other sectors.
jgord 2y ago
have you considered starting / joining a startup instead ?
chithanh 2y ago
> At this point I'm honestly wondering if they aren't just using the interview process to generate high quality validation data for free. Not sure if that is accurate, but one of the reasons why DeepSeek R1 performs so well in certain areas is thought to be access to China's Gaokao (university entrance exam) data.
- karn97 2y ago
  Thats stupid. same as indias IIT advanced for ex. You learn all that stuff in year 1 physics and math in uni.
  - chithanh 2y ago
    Yes, however you also want to distinguish correct from incorrect answers. You get that from the exams, not from year 1 textbooks.

spaceguillotine 2y ago

Bottom is about to drop out thats why, ethics are out the window already and its gonna be worse as they claw to stay relevant. Its a niche product that tried to go mainstream and the general public doesn't want it, just look at iPhone 16 sales and Windows 11, everyone is happier with the last version without AI.

qoez 2y ago

They were always going for regulatory capture. I think deepseek shook them but I don't think we should rewrite the history as them being virtuous only until 2024.