subtech

9 ms·

1. 61.▲
   by subtech 4y ago

   Thanks! Yup, this is basically custom music embeddings + nearest-neighbor vector search. I personally found that vector representations performed significantly better than other approaches. And the results will actually be a lot better once
2. 62.▲
   by subtech 4y ago

   100% ran into a ton of scalability challenges lol. Maybe I should write a blog post about it sometime. But for now, ended up using plain old FAISS.
3. 63.▲
   by subtech 4y ago

   Hmm, not quite. Shazam's approach is more optimized for exact matches. Mine tries to be more optimized for similar matches (if that makes sense lol). So my approach is intentionally a bit fuzzier than something like Shazam, since apart
4. 64.▲
   by subtech 4y ago

   Will 100% improve the UI!
5. 65.▲
   by subtech 4y ago

   Yup sorry about this bug! Need to do some additional post-processing to prevent dupes in the catalog from showing up in the results.
6. 66.▲
   by subtech 4y ago

   Thanks a lot for your feedback! You're right, I should probably add a better song player UI element. Also, as I improve the model & add user accounts, recommendations should hopefully get much better and more personalized!
7. 67.▲
   by subtech 4y ago

   Thanks for your feedback!! Yup, model can definitely be improved (and perhaps I should also add a language filter)! Noted! :)
8. 68.▲
   by subtech 4y ago

   Definitely understand the current model's shortcomings that people have mentioned. I actually think this is largely in part due to how the current model + training process is designed. I have some ideas for improving things on this fro
9. 69.▲
   by subtech 4y ago

   Personally, from using it a lot during development, I found that I kinda developed a sense for which types of songs it'll do really well on, and which types of songs it will sometimes struggle with. But a lot of this should go away wit
10. 70.▲
    by subtech 4y ago

    Haha, thanks! :) Will definitely work to improve the current model!
11. 71.▲
    by subtech 4y ago

    Thanks a lot, definitely have a lot of work to do with improving the model!
12. 72.▲
    by subtech 4y ago

    100% agree with a language filter, working on one (along with a better, improved model!)
13. 73.▲
    by subtech 4y ago

    Oof, honestly didn't expect the surge in traffic lol. It should work, as auto-scaling kicks in, etc. Sorry about that!
14. 74.▲
    by subtech 4y ago

    haha didn't expect many people to guess this!
15. 75.▲
    by subtech 4y ago

    Hey thanks for the feedback! I definitely have a lot of improvement to do on the model, it currently performs better for some styles/genres of music than others. But the model architecture I'm using is kinda outdated as well, gott
16. 76.▲
    Show HN: I trained an AI model on 120M+ songs from iTunes(maroofy.com)

    753 points by subtech 4y ago|427 comments
17. 77.▲
    by subtech 4y ago

    interesting ideas, thanks!
18. 78.▲
    by subtech 4y ago

    yooo thanks! Got some exciting updates shipping super soon (e.g., search, email newsletter, auto-freshing a new list of papers on every page refresh, etc.)
19. 79.▲
    by subtech 4y ago

    lool
20. 80.▲
    by subtech 4y ago

    Yup only abstracts for now, but I think it should be possible to parse the paper PDF to extract just the approach + experiment results w/ GPT-3 (despite 4k token limit). Will add this soon.
21. 81.▲
    by subtech 4y ago

    Ah I think the loading error might be due to many people loading the site at the same time. The site should work now! Newsletter coming, and will definitely add filtering by tags/topics + search! Comments sounds like a good idea as wel
22. 82.▲
    by subtech 4y ago

    Nah, I just built this as a fun personal project.
23. 83.▲
    by subtech 4y ago

    GPT-3! Thank god I don't have to hand write them lol.
24. 84.▲
    by subtech 4y ago

    Yup!
25. 85.▲
    Show HN: ArXiv Feed – An easy way to keep up with AI research(arxiv-feed.vercel.app)

    32 points by subtech 4y ago|17 comments
26. 86.▲
    by subtech 4y ago

    Haha thanks! :)
27. 87.▲
    Show HN: Uminal – Large language model anyone can extend with new apps/skills(uminal.com)

    3 points by subtech 4y ago|2 comments
28. 88.▲
    Uminal: Extend LLMs with custom apps that auto-compose with each other(loom.com)

    1 points by subtech 4y ago|1 comments
29. 89.▲
    by subtech 4y ago

    Hey HN! I'm working on a new project called Uminal, which enables anyone to extend large language models (like GPT-3) with custom apps that auto-compose w/ each other. In short, imagine if anyone could extend LLMs with new skills
30. 90.▲
    by subtech 4y ago

    Currently, it integrates with the Apple Music/iTunes catalog in order to serve the song previews.