[rettichschnidi]

Once again this kind of argument:

> "The reality is that if only a handful of companies and a handful of governments have the resources" to rebuild models, it is not a practical goal for open-source AI.

Any chance one could build a (working) system for training in a distributed way?

Because e.g. Folding@home was able to accumulate quite some computational power this way:

> Folding@home is one of the world's fastest computing systems. With heightened interest in the project as a result of the COVID-19 pandemic,[8] the system achieved a speed of approximately 1.22 exaflops by late March 2020 and reached 2.43 exaflops by April 12, 2020,[9] making it the world's first exaflop computing system. This level of performance from its large-scale computing network has allowed researchers to run computationally costly atomic-level simulations of protein folding thousands of times longer than formerly achieved.

Source: https://en.wikipedia.org/wiki/Folding@home