plusultra on Nostr: There's only one solution: Run a personal, local agent. The billion zillion GPUs are ...

There's only one solution:

Run a personal, local agent.

The billion zillion GPUs are only necessary for training. You can fit a 30B-70B open-source model on a consumer GPU(s).

If people can afford $1,000+ gaming GPUs, then there's no excuse to not have a personal AI workstation.

The latency is also amazing.

quoting
note1hq4…2v69

Signal’s Meredith Whittaker on the privacy nightmares that agentic AI introduces.