Hey. Yeah you. No don’t look over your shoulder. I’m not talking to the guy behind you. Look, we’ve been meaning to tell you that you’re doing a pretty good job out there. Proud of you. Keep up the good work.

  • 0 Posts
  • 91 Comments
Joined 5 months ago
cake
Cake day: November 18th, 2024

help-circle







  • I think that assumes a fundamental difference between the ruling class proletariat that I don’t think exists. Not to the extent that the “proletariat ruling class” would relinquish power more gracefully than the current one.

    I’d say phasing into a system with no ruling class would be the ideal. Yk the whole “once everyone’s super no one is.” Kinda thing.




  • I think the next bit of performance may be leaning hard into QAT. We know there is a lot of wasted precision in models, so the more we understand that during training the better quality small quants can get.

    I also think diffusion LLMs ability to change previous tokens is amazing. As well as the ability to iteratively use an auto regressive LLM to increase output quality.

    I think a mix of QAT and iterative interference will bring the biggest upgrades to local use. It’ll give you a smaller higher quality model thay you can decide to run for even longer for higher quality outputs.