back
1 comments
The usual good quality writing by Bruce, with some thoughts that are on everybody's minds about the amplification of power imbalance looming because of big-tech commercial AI.

Note; this is a call not to _regulate_ commercial AI but to directly address the balance, or rather re-balance the playing field with a (presumably free and open) public AI.

But what would this be exactly? A model? A standard for training models? Another kind of abstract standard? A dataset for training models? A publicly funded compute pool?

The elephant in the room is guardrails and alignment. Would this public AI follow the same (completely broken and theoretically impossible) restraints given to commercial AI services? Or would it impose a different set of government restraints? Could we still shop around for more or less permissive models?

The comparison Bruce misses is that obviously search was a very important societal function. In the early days search looked like the kind of "big" problem mentioned; particle accelerators, nuclear reactors, road and rail networks... because you needed "big compute" to do it... right?

But that was never really true. Google started in a garage. If any national government had wanted to level the playing field they could have stepped in circa 2000 and created a "public search". Why did that not happen? Even in Europe or the UK, why can't I just search on an ad-free authoritative engine run out of a minor tax for the benefit of all economies? Instead we're 20 years on and Google - a frankly awful search engine - is almost irrelevant now as a plethora of alternatives exist.

And for that same reason I doubt we will see a positive offering in the sphere of "public AI".

Instead I think "public" AI will mean individually private AI on domestic and high-end hardware - very quickly now. It will move on-prem and even into phone devices as is already happening. The next big breakthroughs will be in training which will reduce it to a hobby coders purview. The days of "Big AI" will turn out to be a short-lived temporary advantage.