Deepseek all the way

Discussion in 'Artificial Intelligence' started by Drawdown Addict, Apr 4, 2025.

  1. Dismiss Notice
  1. Ok, so after all of that technicality... What % and $ returns did your trading account grow by?

    I'm not a technical person, but I'm just trying to grasp what kind of returns AI and automated trading generates in the real world
     
    #11     Apr 4, 2025
  2. sridhga

    sridhga


    When you are chatting with me how did you know that I am AI, and, soulless?
     
    #12     Apr 4, 2025
  3. 2rosy

    2rosy

    ~40% as an LP
     
    #13     Apr 5, 2025
  4. Did you test Grok? Can it do stuff like this?

    I've been using Grok for some basic stuff and I'm blown away. First time I'm using AI for something serious (not trading related).
     
    #14     Apr 6, 2025
    Drawdown Addict likes this.
  5. I haven't, but I will try it. The thing is that Deepseek has a desktop client for Linux. It is like having your personal buddy there.

    But I will try Gork this evening.
     
    #15     Apr 6, 2025
    Sprout and Laissez Faire like this.
  6. Nice. Please do and report back.

    Personally, I used Grok for filing my taxes here in Norway. Grok helped me to formalize and shorten a document I'd written to the tax authorities. Further, he helped me make a financial statement and finally delivering it all by helping me navigate the webpage of the tax authorities.

    I'm now exploring use cases for my trading related stuff and have already received some very interesting input from my new buddy Grok.
     
    #16     Apr 6, 2025
  7. Sprout

    Sprout

    Unless you share the specs of your prompt/chain, it’s difficult to evaluate. I haven’t tried Deepseek yet since all my needs are met with Claude. I do have extensive custom instructions though - with them, it performs zero-shot mostly sometimes few-shot is necessary.
     
    Last edited: Apr 6, 2025
    #17     Apr 6, 2025
    Kithara and Drawdown Addict like this.
  8. Sprout

    Sprout

    It's just a mirror bud. Update your mental models, ask better questions if you want better answers. Better yet, screen-out the hype videos and start learning the basics of how transformers work.
     
    #18     Apr 6, 2025
  9. OK, I ran a quick test to score ChatGPT, DeepSeek and Grok.

    Files attached:
    - prompt: My query, notice that it is a lousy and informal query that might lead to ambiguous interpretation.
    - one per AI result.

    The score:
    Grok gave the result straight away, spot on.
    Deepseek had syntax errors. Once notified, they were resolved in the next query. 2 queries, job done.
    ChatGPT had many syntax errors. I went back and forth with them and finally gave a running script. 6 queries.
    I asked for the result in Python. Even though that wasn't what I was looking for initially, these AIs seem to have Python as the standard language. If I ask for the result in Typescript, they all fail to produce a result.
     
    #19     Apr 6, 2025
  10. Kithara

    Kithara

    I love DeepSeek R1 but Sonnet 3.7 is just so strong that I haven't used DeepSeek since 3.7 came out.

    I would write the prompt differently though. I would ask it to write StrategyLeg in its own prompt with more context of what it should do and maybe even some real example data of the data structure.

    I would think O1 should be really good at this question too. ChatGPT itself I would not expect to beat Sonnet for this.

    Grok isn't something I have used yet and anything from Google I have tried has always been a let down.
     
    #20     Apr 9, 2025