Claude 3.5 Haiku: Unfiltered Review! The Good, the Bad, and the Ugly

Поделиться
HTML-код
  • Опубликовано: 20 дек 2024

Комментарии •

  • @MrMoonsilver
    @MrMoonsilver Месяц назад +1

    Cool topic! As always.

    • @Codewello
      @Codewello  Месяц назад

      Glad you think so!😁

  • @xlretard
    @xlretard Месяц назад +1

    great content 👍

  • @georgemontgomery1892
    @georgemontgomery1892 Месяц назад +1

    Prompt Cache makes price much better.

    • @Codewello
      @Codewello  Месяц назад +1

      Correct, but it's still 4 times the price of the original model.

  • @codigomovil
    @codigomovil Месяц назад +1

    In my own test, I also noticed better overall coding and design style, and I did notice much lower prices-but not as low as in your examples.
    I do have a long set of custom rules that I found help enhance the accuracy and precision of the answers.
    Do you think that’s the reason I'm getting higher costs?

    • @Codewello
      @Codewello  Месяц назад

      I think you get charged more when you send all of the tokens (code) for edits.

  • @perschistence2651
    @perschistence2651 Месяц назад +1

    I thought they would LOWER the price 20% or so. Instead they increased the pricing 4x

    • @Codewello
      @Codewello  Месяц назад

      They think about profit, I'm afraid they will start getting greedy.

  • @Swooshii-u4e
    @Swooshii-u4e Месяц назад +1

    What's happening with the cats on the super maven video. Google APIs cline projects portfolios etc. I added a few comments over there. It looks like the 5th comment didn't go through. I gotta check if I still have it saved. Google didn't let me paste the entire msg so I had to break it up into pieces

    • @Codewello
      @Codewello  Месяц назад

      are you talking about this comment bro ?
      i just saw it "you can show how you the new Gemini, Claude, deepseek, 2 more models up and running with API working. (Repetition for learning) Then with the remaining models you can just show the results you got from the one time prompt."

    • @Swooshii-u4e
      @Swooshii-u4e Месяц назад

      @@Codewello that is the second part. There was a 3rd and 4th part but I only see the 1st and 2nd go through. I know I pasted the huge comment somewhere. If I find it will break it up into 3 parts and send it to you one per 5min

  • @obetishere9215
    @obetishere9215 Месяц назад +1

    If that is the price I will use mistral large 2 instead

    • @Codewello
      @Codewello  Месяц назад

      Have you tried Mistral Large 2? Is it any good?

    • @obetishere9215
      @obetishere9215 Месяц назад

      @ on par with gpt4o with my use case. I am using claude, openai, mistral and deepseek api for instruction following they are all the same.
      My use case is i provide code structure that i implement to llms, then they should follow it.
      I only use deepseek for exams. I created something like duolingo for my personal consumption but topics depend on what i currently study. Deep seek is good for that use case since you can chain exams into harder scenarios as you progress without worrying about the cost

    • @obetishere9215
      @obetishere9215 Месяц назад

      @ I don’t let the llms generate the code from scratch I feed them the architecture based on my boiler plate. They are inconsistent as of now. Currently llms are good for instruction following

    • @obetishere9215
      @obetishere9215 Месяц назад

      @@Codewello on par with gpt4o with my use case. I am using claude, openai, mistral and deepseek api for instruction following they are all the same.
      My use case is i provide code structure that i implement to llms, then they should follow it.
      I only use deepseek for exams. I created something like duolingo for my personal consumption but topics depend on what i currently study. Deep seek is good for that use case since you can chain exams into harder scenarios as you progress without worrying about the cost

  • @not_White_ninja
    @not_White_ninja Месяц назад +4

    just tell me that you're not an egyptian 😂

    • @Codewello
      @Codewello  Месяц назад +2

      I'll be lying, bro. 🤣

  • @perschistence2651
    @perschistence2651 Месяц назад +1

    In my eyes:
    The good: Coding
    The bad: Pricing, Wokeness and many more refusals than Sonnet.

    • @Codewello
      @Codewello  Месяц назад

      What do you mean by "wokeness"? Is it trained on some agenda?

  • @Codewello
    @Codewello  Месяц назад

    The fun AI experiment predicts the election Results :ruclips.net/video/JObuWC0N5gg/видео.html

  • @isheriff82
    @isheriff82 Месяц назад +1

    But its price is fucked up

    • @Codewello
      @Codewello  Месяц назад

      Totally agree with you; it's the ugly part of it.