Comments by "Chrysippus" (@4.0.4) on "bycloud" channel.

  1. Sure, but what does that have to do with chip manufacturing?
    1
  2. I will own stuff, eat beefy steaks and make them angry.
    1
  3. Noooo you guys don't get it, we need 7 trillion dollars ideally in Sam Altman's creepy eyeball shitcoin, otherwise AGI will take over the woooorld. ...Also wtf was 7:22 LMAO
    1
  4. That 1024px is probably itself an upscale. Dall-E is, for example.
    1
  5. Thanks for putting all the links in the description. MusicGEN, Tracking Everything and MatteAnything are so potentially useful!
    1
  6. Basically, you're one of the many many people who get interested in AI because they want to build a waifu. Don't get me wrong, it's great to have some strong motivation!
    1
  7. Mistral released a 24B model a week ago (yesterday in human time) and I think they'll do the same as DeepSeek, so Thinkstral? Mindstral? Soon. Also, Apache license!
    1
  8. ST can plug into Kobold, not just Ooba; notably, Kobold offers better GGUF inference. There is even a "Frankenstein" version which has a few novel 3-bit and even 2-bit (and now lower) quantization methods (< 2 bit not recommended). If you have a 3090 or 4090, do not use 7b models. You can run so much better ones. Give 8x7b ones a try, such as Noromaid-v0.4-Mixtral-Instruct-8x7b-Zloss, I'd also recommend Crunchy-onion-Q4_K_M.gguf (little known one, but fun).
    1
  9. The modern left is basically the evangelicals of the 1990s that said Harry Potter was satanic. Right now Harry Potter is transphobic against their DEIty or something.
    1
  10. You really have to check for yourself to decide which one you think is best (or works best for you), but the Arena does give you an idea of which ones are the contenders.
    1
  11. Space man bad 🙄
    1
  12. It's strange. When I was a kid I had no money but wanted to buy cool stuff. Now I have money but don't want to give it to evil pro-regulation monopolies.
    1
  13. It seems very convoluted, but I guess it should learn with less data? That could be good for startups that don't have big datasets.
    1
  14. Scenery sketching AI won't work with models that inherently need a low-res full frame of input and output.
    1
  15.  @lilgarbagedisposal9141  you do know langchain has been around for some time right? This is very easy to use, but not some new breakthrough.
    1
  16.  @phen-themoogle7651  I don't get it. OpenAI made CoT easy for the masses (or rather, those with a subscription), but this is nothing more. Sure it's an easier way to burn $0.25/prompt, but am I missing something?
    1
  17.  @PhantomAyz  the economics are very prohibitive, they probably have proofs of concept internally. But I expect them to release this soon now.
    1
  18. Kinda spooky what elites might get access to only behind closed doors if free research for all is hampered by regulation.
    1