11 comments

  • nancy_le a day ago

    How do you plan to handle the variability in video quality and content style on YouTube? For example, some educational channels might have a very structured approach, while others might be more conversational or rambling. How does your AI model adapt to these differences, and what kind of limitations have you encountered so far?

    • rtills a day ago

      Good question - currently it is just hooked into the OpenAI API. I think it actually performs generally well for its purposes but to make it get more traction I would have to consider fine-tuning or prompt engineering further

      The limitations are mostly that the questions and flashcards can veer into being generic - I've still found them useful in my own dogfooding though...

  • HardwareLust 2 days ago

    Pretty impressive so far. How are you paying for the compute on this?

    • rtills 2 days ago

      My own pocket at the moment. Very happy for folks to try and use it as a tool. If it becomes a problem - hopefully the community will contribute. What do you think?

      • HardwareLust 2 days ago

        Hopefully you won't be overwhelmed too quickly. =)

  • codingdave 2 days ago

    How does this give back to the people whose videos are being used?

    • rtills 2 days ago

      Hi - could you explain more? If you look at the site - the videos are embedded on the page so should still generate the same revenue -> https://asterlab.io/

  • toomuchtodo 2 days ago

    Can you support any media target yt-dlp supports?