I recreated Google Gemini's fake demo, but for REAL (GPT-4 Vision) - Try it yourself!

Поделиться
HTML-код
  • Опубликовано: 26 янв 2025

Комментарии • 13

  • @show-me-the-data
    @show-me-the-data  Год назад

    Hey folks! Let me know if you have any questions! Would be happy to tell you more about the process of how this was made! Let me kick things off.
    Q: Will this work with open source vision models like LLAVA?
    A: I tried this out. It's not there yet. It gets confused with the sequence of images and the answers are far from perfect. Wait a few more months.
    Q: Does it understand the context of the entire conversation?
    A: No, only the last question asked. In theory, with a slight tweak to the code, you can provide the previous back and forth chat history so you can have a natural conversation.
    Q: This doesn't work on Browser [Insert Browser]!
    A: Yes, I mentioned in the video I only tried this with Chrome. Feel free to contribute to the repo if you want to add more support!

  • @RolandoLopezNieto
    @RolandoLopezNieto Год назад

    Amazing job buddy, subscribed

  • @CodexPermutatio
    @CodexPermutatio Год назад

    Transparency and code. This is the way! Well done.

  • @MeinDeutschkurs
    @MeinDeutschkurs Год назад

    Gorgeous! You’re so creative and inspiring! Amazing! What the quack! ❤

    • @show-me-the-data
      @show-me-the-data  Год назад

      The people who watch to end will appreciate this more 😅❤

    • @MeinDeutschkurs
      @MeinDeutschkurs Год назад

      ⁠@@show-me-the-data, Certainly! I'm confident that the community can popularize the phrase 'what the quack' in English! 😂 It's a much more comfortable alternative to the version ending with 'F.' I love it and thank you for bringing/picking it up. I’m ready to use it! 🤣🤣

  • @baharalmasi
    @baharalmasi Год назад

    Mind blown🙏👌👌👌

  • @paulevans3060
    @paulevans3060 Год назад

    Question: is there a way to teach GPt4-GPTs a sequence and have it store that information in its Knowledge? say i teach to move red ball to box. GPT4 will store this information, then i test this yellow ball(incorrect colour) and get GPT4 to tell me my movement was incorrect and even stop me be fore i complete the movement of putting the yellow ball in the box?

    • @show-me-the-data
      @show-me-the-data  Год назад +1

      Yes you'd have to describe in the instructions your desired response. This is in-context learning so it only requires a prompt, however if you want a system that does that by default you'd have to fine tune the vision model and you can't do that with GPT-4V but can with open source models like Llava

  • @soudimofrad1794
    @soudimofrad1794 Год назад

    Wow this is so amazing. Why do I have to set the language settings? Doesn’t tts api accept any language automatically?

    • @show-me-the-data
      @show-me-the-data  Год назад

      Ah good catch that's just required for the browser speech recognition because otherwise it only uses English

  • @gnsdgabriel
    @gnsdgabriel Год назад

    There is an error on RUclips. It is reporting 500 views instead of 500k.

    • @show-me-the-data
      @show-me-the-data  Год назад +1

      Ah I've come across this one. The fix is simple! Just gotta send it to all your friends with a threatening letter to share it with all their friends 🤷🏻‍♂️