Local AI: - can see reasoning - faster - will do whatever I want

the reasoning view is the bit i actually miss when i go back to hosted models

What’s your local setup?

What app is the left?

I think my biggest fear with investing (would probably go 2x dgx sparks if I did it) is that it won't keep up with what I need it to do Currently using Fable 5.1 high/xhigh to plan, but since I use @mattpocockuk spec/ticket method, I can delegate to opus 5 (usually medium) to implement since the tickets are narrow So I don't doubt that open source can do the implementation part, but the planning too? I guess I could always buy them locally, return them if it just doesn't do the job

'faster', Tok/sec maybe but if that's all I cared about I'd use Gemini

If API AI was faster I might die of sleep deprivation though. I only go sleep when I know the agent is going to take too long anyways.

- won't actively try and kick you in the balls

Choose uncensored and behold :)

its called jev for anyone who is wondering about this post

I'm realizing more and more that it was a great investment to collect so many 3060s, 3070s, 3080s, and 3090s! They're in boxes now, but I'll dust them off soon!

How exactly it will do whatever u want again?

The thing all depends about your hardware without good hardware you just can’t use a 4-14b model since they are just not good enough without fine tuning which could take a lot of resources and time

What about the classifier?

Deepseek Harness Mobile or PWA?

See but with that, you are going to need a good machine to be able to run a good enough model

can this run on my single 4070

Ok shutup and give me gpu

How is it for coding?

@grok what app is that on the left video

DSH 竟然已经适配了移动端 UI？！

I had this weird idea based on what you have been tweeting, and maybe you are already doing this: A p2p based LLM network. Distributed LLM models that cli tools connect to, and run on peoples own machines, on a server, home pc or anything else. Instead of centralized like openai.