There is nothing wrong (NOTHING. WRONG.) with mandatory transparency requirements to allow academics/alignment-researching-organizations to evaluate the capabilities of systems that will soon be vastly-more-intelligent-and-creative than us (+capable of harming everyone alive).
There is if the transparency requirements is to fill out a form prior to any experimentation, it would gum down the works like nothing else. Especially if random people would be able to request changes.
I have a feeling we have completely opposite beliefs here about AI Existential Risk. "Gumming down the works" is a laughable (and highly-unlikely) outcome compared to 'everything else that might happen without a sense of Basic Due Diligence'.
I'm not going to argue further unless you state clearly your views on AI Existential Risk.
AI is not a matter for state regulation, 50 different sets of rules won't work. Nor is it a matter for federal regulation what people run on their computers. Federal preemption of regulation is the best alternative, but too sensible to make it through Congress. (this time)
The people advocating regulation know little about AI, nothing about litigation or government and have a media/vibes/stupidity & evil-based worldview against intelligence itself, whether artificial or natural. Getting the right answers is a threat to many lies.
What is wrong with his AI legislation proposals and more importantly which legislator is doing a better job. You must have a long list, do share. This is not me being snarky, I am genuinely curious about this topic, and you have opinions.
You know nothing about AI, nothing about litigation, nothing about government, and you have a media/vibes/stupidity & evil-based worldview against intelligence itself, whether artificial or natural.
(so now the onus to defend a long series of evidence-lacking assertions is back on you, where it belongs, instead of the author)
I feel like the phrase "what people run on their computers" was one thing from late last century to roughly now, and another thing entirely from roughly now onwards.
"Very MAGA sources proved crucial. Opposition from those sources was vocally motivated by fear of big tech, and a few specific mundane policy concerns like privacy, protecting children, copyright and protecting creatives, _and potential bias against conservatives._" [emphasis added]
Well, if some red state prods all the big labs into putting in infrastructure to check frontier models for toxic Wokeness, perhaps it won't be too much of a stretch to add tests to detect actively deceptive output, which would be at least one of the paths towards existential risks. A COT from a reasoning model agent containing "power, power, I must have more power!" might also raise an eyebrow...
Note Sores, Thillis...recently saw Mira Mutari crop up again too.
Well, that's one checked off the Imminent Threat List, phew. I wish I could at least file this under Pick Up The Phone, that there's a glimmer of hope for sane regulation (well, less insane regulation) even in the deep red end of the pool. But it seems to have largely been an internecine affair? Outside of the usual concerns about muh copyright law, there seemed to be mostly crickets from Democrats...and, yeah, fair, they had 99 other parts of the Bigly Bill to critique. And the angles of attack are all wrong for building a positive bill-passing coalition, as you note. Notkilleveryoneism trotting out MTG like Harris trotting out Cheney would probably go over equally poorly. So it's unclear what the actual takeaways should be, beyond not completely writing off hope of one's enemies' enemies becoming friends sometimes, somehow.
Zvi, wondering if you will discuss how OpenAI has now been ordered to store data indefinitely (so long "temporary chats"). I assume this will apply to other LLM's as well, though I'm not sure.
Any options or hope for people who are trying to keep their data, ideas or conversations private?
Was about the only good thing in the bill. Sad. Now AI will be smothered by “do-gooders”
There is nothing wrong (NOTHING. WRONG.) with mandatory transparency requirements to allow academics/alignment-researching-organizations to evaluate the capabilities of systems that will soon be vastly-more-intelligent-and-creative than us (+capable of harming everyone alive).
There is if the transparency requirements is to fill out a form prior to any experimentation, it would gum down the works like nothing else. Especially if random people would be able to request changes.
I have a feeling we have completely opposite beliefs here about AI Existential Risk. "Gumming down the works" is a laughable (and highly-unlikely) outcome compared to 'everything else that might happen without a sense of Basic Due Diligence'.
I'm not going to argue further unless you state clearly your views on AI Existential Risk.
My view is that we should accelerate even further and faster, “boldly go where no man has gone before”
AI is not a matter for state regulation, 50 different sets of rules won't work. Nor is it a matter for federal regulation what people run on their computers. Federal preemption of regulation is the best alternative, but too sensible to make it through Congress. (this time)
The people advocating regulation know little about AI, nothing about litigation or government and have a media/vibes/stupidity & evil-based worldview against intelligence itself, whether artificial or natural. Getting the right answers is a threat to many lies.
What about Scott Wiener?
What about that contender for worst entity?
What is wrong with his AI legislation proposals and more importantly which legislator is doing a better job. You must have a long list, do share. This is not me being snarky, I am genuinely curious about this topic, and you have opinions.
I'm going to counter very intelligently with:
You know nothing about AI, nothing about litigation, nothing about government, and you have a media/vibes/stupidity & evil-based worldview against intelligence itself, whether artificial or natural.
(so now the onus to defend a long series of evidence-lacking assertions is back on you, where it belongs, instead of the author)
I feel like the phrase "what people run on their computers" was one thing from late last century to roughly now, and another thing entirely from roughly now onwards.
Podcast episode for this post:
https://open.substack.com/pub/dwatvpodcast/p/ai-moratorium-stripped-from-bbb
"Very MAGA sources proved crucial. Opposition from those sources was vocally motivated by fear of big tech, and a few specific mundane policy concerns like privacy, protecting children, copyright and protecting creatives, _and potential bias against conservatives._" [emphasis added]
Well, if some red state prods all the big labs into putting in infrastructure to check frontier models for toxic Wokeness, perhaps it won't be too much of a stretch to add tests to detect actively deceptive output, which would be at least one of the paths towards existential risks. A COT from a reasoning model agent containing "power, power, I must have more power!" might also raise an eyebrow...
Note Sores, Thillis...recently saw Mira Mutari crop up again too.
Well, that's one checked off the Imminent Threat List, phew. I wish I could at least file this under Pick Up The Phone, that there's a glimmer of hope for sane regulation (well, less insane regulation) even in the deep red end of the pool. But it seems to have largely been an internecine affair? Outside of the usual concerns about muh copyright law, there seemed to be mostly crickets from Democrats...and, yeah, fair, they had 99 other parts of the Bigly Bill to critique. And the angles of attack are all wrong for building a positive bill-passing coalition, as you note. Notkilleveryoneism trotting out MTG like Harris trotting out Cheney would probably go over equally poorly. So it's unclear what the actual takeaways should be, beyond not completely writing off hope of one's enemies' enemies becoming friends sometimes, somehow.
Zvi, wondering if you will discuss how OpenAI has now been ordered to store data indefinitely (so long "temporary chats"). I assume this will apply to other LLM's as well, though I'm not sure.
Any options or hope for people who are trying to keep their data, ideas or conversations private?
It has been discussed previously on some of the weekly AI posts.