Ask HN: Could we pool our machines to run more powerful AI models locally?
Today I remembered the SETI@Home project. A distributed computing project for people to contribute their computer's power to help search for signals of extraterrestrial life.
And then I wondered if it's feasible to run more powerful open source models across multiple machines. Has it been tried before? The meta project for SETI@Home is called BOINC: https://boinc.berkeley.edu but I'm not sure if it focuses on running large open weights models for inference as much as compute in general. I agree with the premise though.