Post

Except... your "ai+ethics" post are very... mild at best? I guess people might complain about not addressing the more pressing issues of fake videos to destabilize elections, as Putin has successfuly done in US / Brazil / EU etc?

I hadn't heard about those particular fake videos, any good articles I can read? Would you say my ai+ethics coverage is so mild that it doesn't protect me being labeled as a breathlessly proselytizing LLM shill?

Just to be clear I'm not personally trying to blame you for anything yet as technologists we have to be aware of what we're doing to the world and right now it feels like working on AI means "accelerating tech in the hands of the powerful" which basically means: building a jail for the rest of us

Avoiding "accelerating tech in the hands of the powerful" is a driving motivation for me - it's why I spend so much effort showing people how to understand and use this stuff, and why I track both openly licensed models that run on personal devices and the falling costs of hosted models

It's also why I've been so celebratory of the increased completion from more vendors - the idea that only a tiny group of (or even just one) organizations could have exclusive control over this kind of technology terrifies me

The problem is not so much the exclusive control, the problem is that if **one bad** actor has access to a very powerful tech, he might do serious damage There is therefore a civilizational question. Should we continue such developments or not? /1

Whether it's "community-based / open-source" or not is beyond the point, really /2

We know how to build LLMs now, and it turns out it's not even particularly difficult: read some papers, scrape the web and spend a few million dollars on GPU compute We can't uninvent them, and now that they're being trained around the world I very much doubt we can legislate them away either

Sure, LLMs are obviously very limited, we will have to go one step further at the very least to get something meaningful (differentiable symbolic reasoning for AGI or whatever) Now we can choose to continue working on this, or to collectively stop A bit like a non-proliferation treaty

I'm a big AGI/ASI skeptic: I don't understand the arguments for how it doesn't cause massive global economic disruption I'd be happy with the LLMs we have now continuing to get incrementally more useful as tools for humans to use

Well sure... but we DON'T want a massive global economic disruption... so should we continue working on all this? or regulate it?