I have a foreboding of an America in my children's or grandchildren's time when awesome technological powers are in the hands of a very few and no one representing the public interest can even grasp the issues.
That's still my fear. I currently still spend around 20% of my time on exploring and developing sovereign AI principles. It's a slow going, especially because I limit myself to "plebware". Which I emulate with the first model Macbook M1 (16GB unified memory, slow npu.) I did work projects with it (langchain pipelines, mineru, and so on) using only open local models. So it's possible, especially now. But... if you want less than half the agentic capability of Claude or GPT [1] in open weight models, you'll still be putting down at least $40k upfront. Versus a $200/mo sponsored Claude 20x subscription.
It's a tough world when -2000% opex inefficiencies are considered to be easily offloaded to fake retail hype come IPO time.
That's still my fear. I currently still spend around 20% of my time on exploring and developing sovereign AI principles. It's a slow going, especially because I limit myself to "plebware". Which I emulate with the first model Macbook M1 (16GB unified memory, slow npu.) I did work projects with it (langchain pipelines, mineru, and so on) using only open local models. So it's possible, especially now. But... if you want less than half the agentic capability of Claude or GPT [1] in open weight models, you'll still be putting down at least $40k upfront. Versus a $200/mo sponsored Claude 20x subscription.
It's a tough world when -2000% opex inefficiencies are considered to be easily offloaded to fake retail hype come IPO time.
(#1466800) ↩