Threading the needle in this post of anthropic has done some bad things for AI governance & the discourse but the actions of this administration are way worse so we need to get a handle on it before stronger models, open or closed, come along soon. https://www.interconnects.ai/p/welcome-to-the-agi-era-of-ai-governance
Many users welcome calls for stronger AI governance before advanced models arrive because it may spark useful conversations on capabilities, while others criticize officials like David Sacks for failing to prepare adequately.
Again it makes me really uncomfortable to write about these topics, but that's how I can tell I'm learning new skills. It is good for people to speak independently on what is coming, as the companies and governments have incentives of their own.
Threading the needle in this post of anthropic has done some bad things for AI governance & the discourse but the actions of this administration are way worse so we need to get a handle on it before stronger models, open or closed, come along soon. https://www.interconnects.ai/p/welcome-to-the-agi-era-of-ai-governance
@natolambert Great stuff, I just finished my reaction piece as well, but the Android X app won't let me post it as it's too long. (old app bug never fixed) So I have to post it as an article, which android won't let you do (never been implemented) so I have to go to my PC first.
The only win from what is happening is that it may force a serious conversation about what capabilities should be engineered into models for the benefit of humanity.
We will probably never get a dual-use, highly capable model that is generally available again.
Other technologies have gone through this transition.
Nuclear has both societal use and military use, and there continues to be advancement in both areas.
There should be no reason the same cannot be true for AI.
@AndrewCurran_ * first reaction piece. It’s always trying to read where we are pointing. The status quo is mostly noise
@natolambert David Sacks was simultaneously the Crypto and AI Czar. Even if he did literally nothing to prepare us all for advanced AI, the least he could have done is pump our bags. I’ll die on this hill.
@natolambert Maybe we didn’t need to try to give ai personalities. Maybe they should have been engineer, marketed, and sold as tools. But non of that really matters now
@natolambert The real governance gap isn't Anthropic's missteps—it's the absence of any binding pre-deployment evaluation framework before open-weight models cross critical capability thresholds.
@natolambert Short, but a nice summary.
@natolambert Current admistration is a million times better for AI than the alternative would be.
Stop bitching about it