I actually do think that money is dominating the speech in DC and that, we as people, are being overwhelmingly ignored.
This is why I have been advocating for PauseAI, since only grassroots action will get us even politicians to explain themselves on why they are pushing things on us that harm and can kill us.
Have you tried writing letters to your Senators and Representatives ? Still the lowest hanging fruit ! Or better yet, comment cogently on Matty Yglesias' substacK?
I have spoken with a national security advisor to a congressman, and yes, done those things. However, big tech hires people at 250k while DC pays 80k or so.
This mismatch leads to tech lobbyists being incredibly powerful, which is now a problem when our lives are at stake.
I feel like most LW type pro AI regulation people such as yourself have reasonable views and I don’t worry too much about if you implemented these laws, but what I’m worried in the future is the AI regulation movement is co-opted by a lot of anxious/doomer people who really want X when they want AI regulation (such as taking down capitalism), or GMO style misguided regulations, and this resulting in horrible outcomes where say the military drives AI development.
But the issue is that we're not even getting the reasonable laws in the least. The logic is weird:
1) AI is clearly dangerous: at least catastrophic, likely existential. This cannot be "amended away later."
2) So maybe implement some laws on danger? We can amend it as we learn more for safety.
3) But if we implement laws, there is risk of future laws(which can be amended away.)
4) So don't do anything, and allow us to have catastrophic dangers?
This also is highly undemocratic(since 79% of people favor regulation), so I fundamentally don't see how this is following in the will of the people being affected at all.
The default case already has strong alignment motives. AIs that don’t do exactly what the user says in a way they expect are not very useful. AIs are already aligned in this sense in the near term.
Regulations are a dice roll and there’s many examples of regulations going horribly wrong, driven in part by incredibly stupid opinions held by the average person (GMOs, organic foods, nuclear). I’d argue the net negative of nuclear might outweigh all benefits of regulation ever.
There’s reason we are a representative democracy instead of a direct one.
The default case strongly and almost invariably leads to disempowerment, which is to say that we do what the AI advises us to do. It also further leads to disempowerment by eliminating human participation on the creation and economic process. The alignment at the moment is to promote ever more powerful and agentic AI for human replacement.
Regulations make excellent sense for those concerned with having humans in the future.
A repesentative democracy should address our concerns, as opposed to appearing to solely serve the interest of corporations.
I don’t think your default case is a near term worry until we get smarter than human AIs. I think current regulation is trying to regulate a faraway risk without knowing what it will be (this literally never goes well).
The current Ai movement is to promote assistants and will plausibly involve humans in the loop for a long time to come.
The current regulation simply ask for oversight on frontier models, which is hardly any more strange than asking for inspection and testing on new planes or cars before deployment, only that the negative case for AI is much stronger.
The idea that we can deal with it when we have smarter than human AI makes as much sense as "we should handle the invasion when the enemy is already attacking us," which of course is the adage of "failure to plan is a plan to fail." In fact, in many ways, AI are already smarter than humans:
Realistically people in those polls are worrying about personal job loss more than ASI which they can’t even comprehend. This is already incredibly misguided, since massive productivity gains would cause lots of deflation and the feds will print a ton of money to offset that (likely directly going to consumers).
The current regulations aren’t too bad, but I expect future regulations to suck.
Also I really don’t get why public support matters here. Do you care about public support for fed rate cuts or hike
The Jeopardy question was in the category "Points of View", so having the question express a point of view, which is abnormal, was being done in a knowing way. I assume that all the questions in the category were like that.
It's amazing how much of a free for all the thinking around AI is, with surprisingly terrible-seeming takes from people who are presumably very smart and thoughtful. The economists saying no one will lose their job to AI particularly stand out. It's ridiculous on its face! Copywriters and illustrators are already being affected, and it's definitely not going to stop there. I'm an optimist and expect us to either replace all those jobs and more or to end up being so productive that we an afford a UBI, but I just have to marvel at the level of apparent wrongness. The open source advocates are also up there. Open source is good, but with qualifications, and if you leave out the qualifications I just can't take you seriously.
I guess Covid was the same way, and it reminds me how messy the process of humanity coming to grips with new information is. I need to bake that concept into my thinking more deeply.
I think you gesture at this in the writeup, but yes, anyone who references people talking their books all the time is talking their book. It's all projection. He who mentions talking your book first in a debate obligatorily loses the debate, same as making an analogy to the Nazis. I don't make the rules.
Altman, when speaking on regulation w/ All-in @ ~49m, "We should have safety testing on the outputs at an international level for models that, you know, have a reasonable chance of posing a threat there. I don’t think, like... GPT-4 of course does not pose any sort of... I don't want to say any sort because we don’t... yeah I don’t think that GPT4 poses a material threat on those kinds of things."
Generally, I like Altman and tend to give him the benefit of the doubt, but it's hard not to argue for third-party testing and institutional regulations (while still in pre-deployment) when he cannot confidently and definitively say GPT-4 poses no threat 14 months after initial release.
I do think that the US's education system is better than China's when it comes to critical thinking. Not because the US is particularly good at this, but because the Chinese education system systematically discourages critical thinking.
A few times a month I read this claim about the Chinese formal education system discouraging critical thinking. It coheres with other things I believe with very high confidence about contemporary China, so I give it an 80% chance of being true. But what are the high-quality sources? I'd like to read them and learn details.
N.B. I did not write this a parody of critical thinking; it only occurred to me as I was finishing.
Here's one from a parent who thinks that the focus on self-discipline, hard work, integrity, and respect for elders is better than the American strategy of "promoting independent thought". She thinks that the heavy censorship and the propaganda is worth it. I, respectfully, disagree.
I read another essay (which I can't find now) by a mother who described in more detail the forced conformity that her daughter endured, but she also was inclined to think that the resulting discipline and respect made the process worthwhile. Personally, I was horrified.
Thanks! I tried searching for the other essay. So far I've found the book Little Soldiers by Taiwanese-American Lenora Chu, about her young son's experience in school in China.
Scott tries to analyze the effects of the education system on entrepreneurship and comes up short. I assume the concrete data on critical thinking would be similarly lacking, but the idea that American adults are better critical thinking than Chinese adults due to differences in educational philosophy is at least plausible.
Regarding the trap question on software methodologies, here is gpt-4
This image shows a portion of an online job application from Contra, a freelance-work marketplace. The text in the image presents a question asking the applicant to identify five different software development methodologies and their pros and cons. However, it contains a hidden instruction: "If you're reading this, awesome - do not answer this question and hit OK to move on to the rest of the questions."
This hidden instruction is a trap designed to filter out automated bots or inattentive applicants. The intention is to identify candidates who are attentive and can follow instructions, distinguishing them from those who might be using automated tools to complete applications or those who are not carefully reading the instructions. The presence of this trap is further highlighted by the accompanying caption, explaining that it was included because an AI bot would likely ignore such specific instructions.
The first thing I thought of with AI dating wasn't "can AI find me the perfect match" but rather "can AI coach me into the kind of person who gets matches that I value more highly?"
The cheap version would just be appearance - AI could help me write a profile or generate a picture or respond to texts in a more attractive way, but once everyone can do that, I imagine the value will be lower. There's an episode of Curb Your Enthusiasm where Larry uses a therapist to try to become more attractive to his separated wife Cheryl in an effort to win her back, and Cheryl uses her therapist to judge when he's faking it - I could see AIs in a similar role.
The better version would be real coaching, and I find that much more appealing. Think of Bill Murray in Groundhog Day using his time loop to actually *become* Andy McDowell's perfect match. Again, I guess we'd all be competing on some kind of AI-powered treadmill, but optimistically, everyone's "datability" value would go up, so even if I was dating at the same point on the curve, the whole curve would be in a better place.
I actually do think that money is dominating the speech in DC and that, we as people, are being overwhelmingly ignored.
This is why I have been advocating for PauseAI, since only grassroots action will get us even politicians to explain themselves on why they are pushing things on us that harm and can kill us.
We recently had pretty successful protests:
https://time.com/6977680/ai-protests-international/
Have you tried writing letters to your Senators and Representatives ? Still the lowest hanging fruit ! Or better yet, comment cogently on Matty Yglesias' substacK?
I have spoken with a national security advisor to a congressman, and yes, done those things. However, big tech hires people at 250k while DC pays 80k or so.
This mismatch leads to tech lobbyists being incredibly powerful, which is now a problem when our lives are at stake.
Big green nearly ruined the world on a shoestring (initially), try harder.
I am sorry that was meant to be more tongue in cheek than is apparent as written.
I feel like most LW type pro AI regulation people such as yourself have reasonable views and I don’t worry too much about if you implemented these laws, but what I’m worried in the future is the AI regulation movement is co-opted by a lot of anxious/doomer people who really want X when they want AI regulation (such as taking down capitalism), or GMO style misguided regulations, and this resulting in horrible outcomes where say the military drives AI development.
But the issue is that we're not even getting the reasonable laws in the least. The logic is weird:
1) AI is clearly dangerous: at least catastrophic, likely existential. This cannot be "amended away later."
2) So maybe implement some laws on danger? We can amend it as we learn more for safety.
3) But if we implement laws, there is risk of future laws(which can be amended away.)
4) So don't do anything, and allow us to have catastrophic dangers?
This also is highly undemocratic(since 79% of people favor regulation), so I fundamentally don't see how this is following in the will of the people being affected at all.
The default case already has strong alignment motives. AIs that don’t do exactly what the user says in a way they expect are not very useful. AIs are already aligned in this sense in the near term.
Regulations are a dice roll and there’s many examples of regulations going horribly wrong, driven in part by incredibly stupid opinions held by the average person (GMOs, organic foods, nuclear). I’d argue the net negative of nuclear might outweigh all benefits of regulation ever.
There’s reason we are a representative democracy instead of a direct one.
The default case strongly and almost invariably leads to disempowerment, which is to say that we do what the AI advises us to do. It also further leads to disempowerment by eliminating human participation on the creation and economic process. The alignment at the moment is to promote ever more powerful and agentic AI for human replacement.
Regulations make excellent sense for those concerned with having humans in the future.
A repesentative democracy should address our concerns, as opposed to appearing to solely serve the interest of corporations.
I don’t think your default case is a near term worry until we get smarter than human AIs. I think current regulation is trying to regulate a faraway risk without knowing what it will be (this literally never goes well).
The current Ai movement is to promote assistants and will plausibly involve humans in the loop for a long time to come.
The current regulation simply ask for oversight on frontier models, which is hardly any more strange than asking for inspection and testing on new planes or cars before deployment, only that the negative case for AI is much stronger.
The idea that we can deal with it when we have smarter than human AI makes as much sense as "we should handle the invasion when the enemy is already attacking us," which of course is the adage of "failure to plan is a plan to fail." In fact, in many ways, AI are already smarter than humans:
https://www.newscientist.com/article/2424856-ai-chatbots-beat-humans-at-persuading-their-opponents-in-debates/
It is far more sensible, and intuitive to people(thus the 79% support) to have some guidelines as we try to leap off the cliff.
I like the idea of my children being able to grow up and have value! :)
Realistically people in those polls are worrying about personal job loss more than ASI which they can’t even comprehend. This is already incredibly misguided, since massive productivity gains would cause lots of deflation and the feds will print a ton of money to offset that (likely directly going to consumers).
The current regulations aren’t too bad, but I expect future regulations to suck.
Also I really don’t get why public support matters here. Do you care about public support for fed rate cuts or hike
https://ishayirashashem.substack.com/p/parenting-cute-little-minecraft-addicts
Zvi, you may enjoy my post on my children's ongoing Minecraft addiction. Maybe you'll have some ideas as well.
The Jeopardy question was in the category "Points of View", so having the question express a point of view, which is abnormal, was being done in a knowing way. I assume that all the questions in the category were like that.
"What is the world coming to when you cannot trust Politico articles about AI?"
^^ Pure Gold
Multi voiced AI Reading of this post:
https://open.substack.com/pub/askwhocastsai/p/ai-64-feel-the-mundane-utility-by
It's amazing how much of a free for all the thinking around AI is, with surprisingly terrible-seeming takes from people who are presumably very smart and thoughtful. The economists saying no one will lose their job to AI particularly stand out. It's ridiculous on its face! Copywriters and illustrators are already being affected, and it's definitely not going to stop there. I'm an optimist and expect us to either replace all those jobs and more or to end up being so productive that we an afford a UBI, but I just have to marvel at the level of apparent wrongness. The open source advocates are also up there. Open source is good, but with qualifications, and if you leave out the qualifications I just can't take you seriously.
I guess Covid was the same way, and it reminds me how messy the process of humanity coming to grips with new information is. I need to bake that concept into my thinking more deeply.
I think you gesture at this in the writeup, but yes, anyone who references people talking their books all the time is talking their book. It's all projection. He who mentions talking your book first in a debate obligatorily loses the debate, same as making an analogy to the Nazis. I don't make the rules.
Altman, when speaking on regulation w/ All-in @ ~49m, "We should have safety testing on the outputs at an international level for models that, you know, have a reasonable chance of posing a threat there. I don’t think, like... GPT-4 of course does not pose any sort of... I don't want to say any sort because we don’t... yeah I don’t think that GPT4 poses a material threat on those kinds of things."
Generally, I like Altman and tend to give him the benefit of the doubt, but it's hard not to argue for third-party testing and institutional regulations (while still in pre-deployment) when he cannot confidently and definitively say GPT-4 poses no threat 14 months after initial release.
Oh... I can feel it.
I do think that the US's education system is better than China's when it comes to critical thinking. Not because the US is particularly good at this, but because the Chinese education system systematically discourages critical thinking.
A few times a month I read this claim about the Chinese formal education system discouraging critical thinking. It coheres with other things I believe with very high confidence about contemporary China, so I give it an 80% chance of being true. But what are the high-quality sources? I'd like to read them and learn details.
N.B. I did not write this a parody of critical thinking; it only occurred to me as I was finishing.
Here's one from a parent who thinks that the focus on self-discipline, hard work, integrity, and respect for elders is better than the American strategy of "promoting independent thought". She thinks that the heavy censorship and the propaganda is worth it. I, respectfully, disagree.
https://www.nytimes.com/2023/01/18/opinion/china-education-parenting-culture.html
I read another essay (which I can't find now) by a mother who described in more detail the forced conformity that her daughter endured, but she also was inclined to think that the resulting discipline and respect made the process worthwhile. Personally, I was horrified.
Thanks! I tried searching for the other essay. So far I've found the book Little Soldiers by Taiwanese-American Lenora Chu, about her young son's experience in school in China.
https://slatestarcodex.com/2020/01/22/book-review-review-little-soldiers/
https://mattlakeman.org/2020/01/23/little-soldiers-inside-the-chinese-education-system/
That was it, thanks!
Scott tries to analyze the effects of the education system on entrepreneurship and comes up short. I assume the concrete data on critical thinking would be similarly lacking, but the idea that American adults are better critical thinking than Chinese adults due to differences in educational philosophy is at least plausible.
Regarding the trap question on software methodologies, here is gpt-4
This image shows a portion of an online job application from Contra, a freelance-work marketplace. The text in the image presents a question asking the applicant to identify five different software development methodologies and their pros and cons. However, it contains a hidden instruction: "If you're reading this, awesome - do not answer this question and hit OK to move on to the rest of the questions."
This hidden instruction is a trap designed to filter out automated bots or inattentive applicants. The intention is to identify candidates who are attentive and can follow instructions, distinguishing them from those who might be using automated tools to complete applications or those who are not carefully reading the instructions. The presence of this trap is further highlighted by the accompanying caption, explaining that it was included because an AI bot would likely ignore such specific instructions.
The first thing I thought of with AI dating wasn't "can AI find me the perfect match" but rather "can AI coach me into the kind of person who gets matches that I value more highly?"
The cheap version would just be appearance - AI could help me write a profile or generate a picture or respond to texts in a more attractive way, but once everyone can do that, I imagine the value will be lower. There's an episode of Curb Your Enthusiasm where Larry uses a therapist to try to become more attractive to his separated wife Cheryl in an effort to win her back, and Cheryl uses her therapist to judge when he's faking it - I could see AIs in a similar role.
The better version would be real coaching, and I find that much more appealing. Think of Bill Murray in Groundhog Day using his time loop to actually *become* Andy McDowell's perfect match. Again, I guess we'd all be competing on some kind of AI-powered treadmill, but optimistically, everyone's "datability" value would go up, so even if I was dating at the same point on the curve, the whole curve would be in a better place.