A quick thought on this whole "control of AI models" debate.
Handguns and Fentanyl are available all over the world, but we still don't put them in the convenience aisle at grocery stores.
The reason we don't is because we want friction between people having negative, impulsive thoughts and them actually taking those actions.
If ChatGPT, Claude, and the most popular open source models answer every question about harming oneself, or hurting someone else, or hacking into X or Y, or making a virus that only kills yapping chihuahuas, we do not want that AI to help with that. We want it to say something like, "No, sorry, I can't help with that. But hey let's talk about it, what's going on?"
That's a control. Same for planning terror activity. Or to kill your spouse.
Yes, models will be available that help you with those things. But they shouldn't be the norm, for billions of people on Earth.
This is common sense, but it's somehow being spun into an evil control narrative.
Are there rich people who'd like to control all the smart AI and make everyone use the dumb ones? Sure. Are there governments who want to keep the people down by not giving them all the information? Sure. 100%.
That doesn't mean we scrap our existing laws, or stop making new ones, just because some people want to abuse them.
So you know where I'm coming from: my whole purpose in all of this is making sure everyone on Earth has the best AI, with open models that are as good as or better than anything closed-source. No AI poverty line, where the rich get the smart models and everyone else gets the dumb ones. It's the same reason I'm building SAFE.
The solution to bad science is better science, not alchemy or mysticism. The solution to bad democracy is better democracy, not authoritarianism or anarchy.