oleganza
Joined 18 years ago
Joined 18 years ago
Of course you'd have compliance department to look over the processes to minimize the risk of mistakes and correct them ASAP (if you don't, you'd be liable for negligence and open yourself to a higher chance of failures). But since mistakes and miscommunication is inevitable anyway, you'd require hard limits in every obligation - in terms of time, money, scope etc.
Organizations that are not protecting themselves do not live long. And this principle has nothing to do with the nature of their activity.
Also: releasing just the weights is like releasing some source code without makefiles or configuration scripts. Sounds half-baked.
PS. The real problem with concentration of mega-models is the same old problem of temporary monopolies in emergent markets (IBM in 80s, Microsoft in 90s, Google Search in 2000s). Let's not worry too much about other peoples' money and use the amazing tools we have to solve our problems today.
The author of the article also demonstrated a preference: according to certain criteria of moral obligation they decided for themselves that it's better to "do something because you feel like doing it, not because you feel like you have to."
Again, to manifest such preference you have to be the main character and not an NPC.
Only non-NPC can decide to what degree that want to be like an NPC and what meaning they put into it. NPCs don't have ability to put meaning into things or display preference.
Or even better, even formulating the deterministic theory would require 1000000x energy than approximation. And that's how'd we learn that biological minds do the same trick and skip the hard work :-)
I think what'd be a stronger point is talking about centralization of the quality models. Modern AI tools are inherently centralized around huge shared infrastructure that gives enormous leverage (== capacity for abuse) to those owning the infrastructure. This is true even if you have strong competition among several players: each of them would converge on some business model and majority of users would not be bothered with long-term consequences if they receive very tangible short-term value.
The tooling is amazing, amount of productivity we unlock is fantastic and it's getting better by the day. But we need to watch out for collateral damage too. The future is somewhere there, but we can steer it towards being more or less hazardous.
Yet, those who opt in do have a different opinion. We got two a decade ago, and then a couple years ago through of FOMO that when we are 45 we'd look back and regret missing the window of having another couple of kids. So we did. I'm 39, have four kids, had to get a bigger car, pay the airline tickets through the nose, spend a lot of time on kids' stuff, and love it. My family is the center of the universe and I'm the happiest and wisest dad alive. Everyone else is childish ;-P
(Father of 4, 39 y.o., non-religious.)
The transportation costs are annoying, but worth it.
Web: http://oleganza.com
Twitter: @oleganza
Email: [email protected]
[ my public key: https://keybase.io/oleganza; my proof: https://keybase.io/oleganza/sigs/AJQSAr3c4OGeX1kGNlkC7CA4GCHrwFxd_d0NRObC-P0 ]