Hacker News new | past | comments | ask | show | jobs | submit
formerly_proven | on: Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows
4K bucks buys you around 180 months of <insert AI subscription here> with zero upfront cost.
zamalek | parent | next
Problem is that might go away or get nerfed.
ody4242 | root | parent | next
then you switch provider, it's not a monopoly
formerly_proven | root | parent
If that happens you can still buy hardware later with almost certainly more (tok/s)/$ and better capabilities to run newer models more efficiently (remember native MXFP4?). Right now basically every generation of accelerator is adding new capabilities. These aren't yearly DirectX 9.0c-compatible GPU performance bumps.

As an individual, for average privacy needs (e.g. open source or at-home coding and automation), it's pretty much complete nonsense financially to self-host LLMs currently or select hardware now based on the capability to do so, and pay thousands of bucks extra.

skohan | parent | next
If you don't mind exfiltrating all your IP to the API provider
Mistletoe | parent
Haha wow. I’m trying to even imagine the AI landscape in 15 years and I can’t.
butlike | root | parent
Instead of saying "I have a MBP with 64gb of RAM" you'll hear people say: "I'm subscribed to Model 9.x11B" and others will comment: "Oh dang, that's a nice model!"