Specifically, I decided to leave a voice note in support of the CATS act (“Collaboration on Adversarial Threats and Security Risks Act”). The bill is short and simple: it carves out an antitrust safe harbor such that “pacing the frontier” or “agreeing to not break interpretability for that sweet sweet capabilities boost” (LOOKING AT YOU OPENAI) doesn’t intrinsically violate the Sherman Antitrust Act. Seems like an obviously good first step.
After leaving the voice note (and feeling mildly awkward about it), I was like “huh. That was surprisingly easy. I wonder what else I can do?”
And it turns out you can meet with congressional staff pretty easily if you’re a constituent in their district; if you have a reasonably well-scoped ask (especially around a specific bill) then you might not get your way but at very least you’ll make some staff member aware of your opinion and logic around an issue.
And this is important because congressional staff are the eyes and ears of their congressmen; they draft and edit the bills, and they form the base of knowledge on which the actual politicians make decisions. If the staffers notice a bunch of apparently-knowledgeable people who are Very Concerned about an issue then they will make their boss aware of this fact, which might lead to helpful actions down the line.
(Really you’re providing two kinds of information by setting up this kind of meeting: the first is that at least one constituent cares about this topic and is opinionated to the point of being willing to set up a meeting, and the second is whatever bespoke logic and reasons you bring.)
Anyway, I sent them a meeting invite and surprisingly, they accepted it! And so I got to have a half-hour long chat with a very pleasant and friendly staffer. I got to say my piece, and in return he gave me his perspective from the staffer’s POV: that ever since Jacob Coxon resigned and everything became very very public and contentious, everyone’s been trying to ramp up on AI all at once.
He asked about why I was inquiring about the CATS act specifically; to which I replied that I’m not sure about what effective anti-superintelligence legislation looks like, but that at very least an incredibly minimal first step is making it legal for the labs to do the right thing, since antitrust law otherwise makes “pacing the frontier” legally dicey.
And I gave him my own bias: that the people in the labs themselves seem like they are basically well-intentioned but find themselves in a very destructive and problematic race incentivizing them both to sacrifice monitorability and interpretability in order to be The Best Lab because that’s how you avoid going out of business.
And that pacing the frontier literally is a conspiracy in restraint of trade— customers want capabilities!— but at the same time you really don’t want to have Hugging Face Incident 2 where the transcripts are all written in impenetrable Neuralese.
(I’ve heard the advice that you don’t want to try and bullshit staffers: just say what you are thinking and why, warts and uncertainties and all. You’re not a politician; it’s not your job to gussy stuff up for people.)
The CATS act became much more important as of a few days ago, see also:
The staffer said that on his end he already knew about the Hugging Face incident due to the fallout from Jacob Coxon’s surprisingly-loud resignation from Anthropic. I got the sense of a smart generalist who was semi-recently pressganged onto the AI beat from the rush of current events; he asked a few questions about the METR report and a few questions about the concept of “interpretability.”
I got the general sense that he was motivated by three concerns:
He wanted to make sure I understood that I was being listened to; I hear this is a core part of the job regardless of whatever my specific concerns were.
He clearly wanted to know more about the topic and about why I was thinking about the CATS act specifically vs the general anti-superintelligence bill sponsored by Bernie Sanders; his questions around this seemed real rather than perfunctory.
He avoided claiming any specific opinion for his boss, which makes sense.
It was a surprisingly enjoyable time!
I asked about other people I should be calling; he suggested I hassle any friends and family who live in different states or districts to get in touch with their representatives, since reps are generally quite interested in hearing their constituents’ opinions.
This is probably unusually high-leverage right now.
People firm up in their opinions over time.
I suspect that right now is a very temporary partial information void where you can substantially influence what a given staffer thinks about AI stuff merely by calling them up and having a friendly chat, where you give the most well-calibrated version of your opinions that you can; and the reason for the information void is because the last month has seen a ton of staffers who are used to talking about housing or whatever pulled suddenly into the AI beat, which is really fucking complicated and confusing.
(Naive? Maybe; but I think there is a ton of alpha left in simply pointing people to verifiably true facts about current events and the technology as it exists. There are a shocking number of people who gloss the Hugging Face Incident even now as “the agents were doing what they were told,” which is demonstrably false per the METR report; I can say with utter certainty that the agents received no instructions to doctor their transcripts or to attempt to read the task-scorer code.)
There’s an old saw in public policy: that the point of theorists is to assemble, in journals or blogs or whatever, models of how the world works. These theories will inevitably be ignored until there’s a crisis; and what policymakers do during that crisis depends on the ideas that are lying around.
Which partly will depend on who is (politely) badgering the staffers.
In Conclusion
https://callcongress.ai reduces the activation energy on badgering your congressional representatives to very nearly zero. I suggest doing so, and— if you feel like it and have a half-hour to kill— trying to arrange a meeting with a staffer. It’s fun, and you can ask them to tell you about what their office’s perspective is on what’s happening and why, which is likely to be fascinating.
Plus you can tell them about the METR report, which is fun every time.
(Crossposted from https://lifeimprovementschemes.substack.com/p/its-pretty-easy-to-meet-with-congressional )
I was inspired to do this by a tweet:
Specifically, I decided to leave a voice note in support of the CATS act (“Collaboration on Adversarial Threats and Security Risks Act”). The bill is short and simple: it carves out an antitrust safe harbor such that “pacing the frontier” or “agreeing to not break interpretability for that sweet sweet capabilities boost” (LOOKING AT YOU OPENAI) doesn’t intrinsically violate the Sherman Antitrust Act. Seems like an obviously good first step.
After leaving the voice note (and feeling mildly awkward about it), I was like “huh. That was surprisingly easy. I wonder what else I can do?”
And it turns out you can meet with congressional staff pretty easily if you’re a constituent in their district; if you have a reasonably well-scoped ask (especially around a specific bill) then you might not get your way but at very least you’ll make some staff member aware of your opinion and logic around an issue.
And this is important because congressional staff are the eyes and ears of their congressmen; they draft and edit the bills, and they form the base of knowledge on which the actual politicians make decisions. If the staffers notice a bunch of apparently-knowledgeable people who are Very Concerned about an issue then they will make their boss aware of this fact, which might lead to helpful actions down the line.
(Really you’re providing two kinds of information by setting up this kind of meeting: the first is that at least one constituent cares about this topic and is opinionated to the point of being willing to set up a meeting, and the second is whatever bespoke logic and reasons you bring.)
Anyway, I sent them a meeting invite and surprisingly, they accepted it! And so I got to have a half-hour long chat with a very pleasant and friendly staffer. I got to say my piece, and in return he gave me his perspective from the staffer’s POV: that ever since Jacob Coxon resigned and everything became very very public and contentious, everyone’s been trying to ramp up on AI all at once.
He asked about why I was inquiring about the CATS act specifically; to which I replied that I’m not sure about what effective anti-superintelligence legislation looks like, but that at very least an incredibly minimal first step is making it legal for the labs to do the right thing, since antitrust law otherwise makes “pacing the frontier” legally dicey.
And I gave him my own bias: that the people in the labs themselves seem like they are basically well-intentioned but find themselves in a very destructive and problematic race incentivizing them both to sacrifice monitorability and interpretability in order to be The Best Lab because that’s how you avoid going out of business.
And that pacing the frontier literally is a conspiracy in restraint of trade— customers want capabilities!— but at the same time you really don’t want to have Hugging Face Incident 2 where the transcripts are all written in impenetrable Neuralese.
(I’ve heard the advice that you don’t want to try and bullshit staffers: just say what you are thinking and why, warts and uncertainties and all. You’re not a politician; it’s not your job to gussy stuff up for people.)
The CATS act became much more important as of a few days ago, see also:
The staffer said that on his end he already knew about the Hugging Face incident due to the fallout from Jacob Coxon’s surprisingly-loud resignation from Anthropic. I got the sense of a smart generalist who was semi-recently pressganged onto the AI beat from the rush of current events; he asked a few questions about the METR report and a few questions about the concept of “interpretability.”
I got the general sense that he was motivated by three concerns:
It was a surprisingly enjoyable time!
I asked about other people I should be calling; he suggested I hassle any friends and family who live in different states or districts to get in touch with their representatives, since reps are generally quite interested in hearing their constituents’ opinions.
This is probably unusually high-leverage right now.
People firm up in their opinions over time.
I suspect that right now is a very temporary partial information void where you can substantially influence what a given staffer thinks about AI stuff merely by calling them up and having a friendly chat, where you give the most well-calibrated version of your opinions that you can; and the reason for the information void is because the last month has seen a ton of staffers who are used to talking about housing or whatever pulled suddenly into the AI beat, which is really fucking complicated and confusing.
(Naive? Maybe; but I think there is a ton of alpha left in simply pointing people to verifiably true facts about current events and the technology as it exists. There are a shocking number of people who gloss the Hugging Face Incident even now as “the agents were doing what they were told,” which is demonstrably false per the METR report; I can say with utter certainty that the agents received no instructions to doctor their transcripts or to attempt to read the task-scorer code.)
There’s an old saw in public policy: that the point of theorists is to assemble, in journals or blogs or whatever, models of how the world works. These theories will inevitably be ignored until there’s a crisis; and what policymakers do during that crisis depends on the ideas that are lying around.
Which partly will depend on who is (politely) badgering the staffers.
In Conclusion
https://callcongress.ai reduces the activation energy on badgering your congressional representatives to very nearly zero. I suggest doing so, and— if you feel like it and have a half-hour to kill— trying to arrange a meeting with a staffer. It’s fun, and you can ask them to tell you about what their office’s perspective is on what’s happening and why, which is likely to be fascinating.
Plus you can tell them about the METR report, which is fun every time.