• 1 Post
  • 210 Comments
Joined 2 years ago
cake
Cake day: October 1st, 2023

help-circle




  • Ive not looked into it so I don’t know what kind of challenges they face. Theoretically, I don’t see where the problem is though…

    The primary input is a users “wishlist” of things they want. Each thing is then compared against a master list which confirms it exists and when it should be available (metadata). This is optional, but offers a more rich experience. Lastly, each thing is queried against a torrent index to try and find it. Its a relatively simple procedure. I guess the only question is whether books appear on these indices or not.

    After a quick glance at the notice on their site, it seems metadata was the problem… or more precisely, no work was being done to move to a new provider. It kinda reads like they lost steam and stopped developing it.







  • I don’t get why they’d be called hallucinations thought. What LM’s do is predict the next word(s). If it hasn’t trained on enough data sets, the prediction confidence will be low. Their whole output is a hallucination based on speculation. If they actually don’t know the next word order, they’ll start spewing nonsense. Though I guess that would only happen if they were forced to generate text indefinitely… at some point they’d cease making (human) sense.

    LMs aren’t smart, they don’t think, they’re not really AI. There aren’t errors, there aren’t hallucinations, this is by design.










  • I have a theory that if everything was pixel perfect, centered, perfectly aligned and looked the same, the thing would look too sterile. There’s basically a perfect world, written down in books and texts that is being taught to students and there’s the real world. In many areas, these two do not match and the above image is the result of someone’s text book world view not matching the real world.

    Could the discover store have a better UI? Yes. Will a centered, down-anchored, pixel perfect button make it better? Subjective.