Running Lightcone Infrastructure, which runs LessWrong and Lighthaven.space. You can reach me at habryka@lesswrong.com.
(I have signed no contracts or agreements whose existence I cannot mention, which I am mentioning here as a canary)
habryka
Karma: 64,821
We already serve a specialized API for agents that agents get pointed to when they make a request, and we don’t block requests to our general graphQL API. But the agents don’t try that when they run into problems.
God, we have tried so many things to keep LW accessible to the LLMs, but the bots and crawlers are so aggressive these days that it’s actually genuinely hard. We currently have some bot protections on because without those we go down immediately as we get hammered extremely aggressively and go down within minutes. I will nevertheless prioritize doing better on this next week. There must be some way to at least serve some kind of cached version of content.
I was trying to convey that it’s stronger to operate in our desired frame instead of the opponent’s. Explaining feels like it can slide into a defensive posture.
I think this very instinct is probably at the core of like 50% of the dysfunction of modern discourse. It’s just people repeatedly insisting to have everything play out in their specific frame, and repeatedly trying to redirect everything towards that. I have a similar deep distaste for what people like to call “media training” the central lesson of which is to “not answer any question asked straightforwardly but instead use it as a jumping off point for talking about the things you want to talk about”. “Media trained” people are obnoxious to talk to, and make communication and understanding of an issue reliably worse.
If the goal is public persuasion, I’d redirect to what we want people talking about, ideally going on the offensive
No, this is bad. Talk about what will help people understand the situation better. If you need to redirect, first answer the question that is being asked of you straightforwardly, even if it seems to you in bad faith, then explain something in return that you think is relevant. Conversation, including public conversation, is a give and take of different perspectives.
From my experience, the general public will not necessarily internalize the “correct” response over time, no matter how valid the argument (hence the illusory truth effect, though I’d want to read more research on this because I suspect on a longer time horizon you may be right)
People will definitely never all become world experts, but I do think good public communication requires an intention to help the audience genuinely understand. This is not always possible, but the intention is where approximately all the good things in public discourse come from.
People definitely cannot tell if those are AI generated if they are driven by a bunch of human judgement. Sure, if you just have your LLM generate the whole thing end to end then people can tell, but that’s not what I am talking about. You can still experience massive uplift if you stay in the loop and steer it in a way that isn’t really true for writing or direct media generation.
I would not describe the communication policies of unions to have helped the world orient better to what unions are doing, as such, I wouldn’t consider the lessons from that world to be good lessons to adopt. This also just really doesn’t match my model of how discourse has played out over the last decade.
Here is the advice I would actually give as someone who has thought about public discourse for much of my professional life:
Your central job is to explain what is going on. Explain the same thing over and over and over again. Lead with understanding and patience. Within a single battle, explaining something repeatedly can often cause the “accusations” to propagate further, but people over time do actually internalize the correct responses and come to associate an argument with its counterargument. This takes time, have patience. Respond consistently with the highest quality counterarguments whenever you see someone making a bad accusation or attack.
Concede only what is actually fair to concede. Many of the worst blunders I’ve seen in the last decade were the result of people being overly conciliatory because they overestimated a threat and thought they should concede more to diffuse it. Mobs in-particular will often try to smell for blood based on political pressure. Giving into an unjustified threat often opens the floodgates to many more attackers as they see that their aggression is getting traction.
Get your best informed people in front of the cameras. Do not hope that just because someone is sympathetic, or charismatic, or has a good social background that they will make a good spokesperson. They will end up saying really dumb stuff, and then you have do paddle back, or be caught in a really awkward position. In the end, the people you trust to actually say good things will probably also be the people other people will trust to say good things. You will deeply regret basically every time you empower an uninformed spokesperson.
Quote tweeting is very high variance. Quote tweeting is the central way to dunk on people, and sometimes some things are worth dunking on. Seeing that a tweet has many more quotes than likes is among the standard ways to tell that a tweet is being badly received (also known as “getting ratioed”). I recommend mostly avoiding it because it’s bad for discourse, but in contrast to the OP here, it certainly seems to be working as a way to spread counter-narratives to things.
I agree that “silence doesn’t work” and “going after people’s personal lives”, though I think the latter is because it’s bad for discourse, not because it doesn’t work. It does unfortunately often work and often needs to be rebutted.
IMO there is a large difference between LLM-assisted video editing, and LLM-assisted video generation. The former is more similar to coding, in that indeed everyone now uses LLMs to code, and this doesn’t generally trigger people’s slop-senses.
Around 8 years ago Dario Amodei, Jan Leike and a few others predicted that we’d probably have AGI by now.
Do you have a link for this? It really matters whether they were like “it’s really quite plausible we will have AGI by now” (which aged IMO very well) or “we will definitely have AGI by now” (which would have aged relatively badly).
We are definitely still somewhat funding constrained, but much less so than we were when this post was written. It’s hard to put into quantitative terms, but I would say money is around half as valuable to us now than it was a few months ago.
This is because they’ve noticed the labs start to implement somewhat sophisticated control mechanisms
This seems quite unlikely to me based on anything I would consider “sophisticated”? I am surprised Buck would say this. Maybe he means something very generous by “start”?
What good thing do you think happens if they attack him?
I mean, in the limit he would get impeached, and at the very least people like him get elected less in the future.
The goal is to get him bought in.
That’s one of many goals. I think it’s very reasonable to critique that this goal trades off too much against other valuable goals (like having reasonable incentives on anyone in power).
Just for common knowledge, while Dustin is no longer blocking funding to rationalist-adjacent organizations, CG has still rejected funding Lightcone Infrastructure (despite Dustin actively wanting to make a grant to us[1]). I find it frustrating to say that “funding isn’t really a bottleneck” while you continue to block funding to a lot of the historically most effective organizations (by the lights of your own staff!).
Funding continues to be a non-trivial bottleneck for many of the best organizations. I continue to be highly concerned about the pressures and selection effects that apply when an organization is trying to get funding from CG.
- ^
Which he indeed ended up doing directly from Good Ventures after (as far as I know) CG-leadership vetoed grant recommendations by its staff to us (and which I am very grateful to Dustin for doing).
- ^
This might not get around antitrust concerns. At least when I read about this a while ago, my strong sense was that the way antitrust law is phrased is that it would require specifically a US regulator to put restrictions on US companies.
The standard reason why anything would probably add a lot on EU Code of Practice enforcement is that IIRC, most judgements by the EU take many years to play out, often decades. IIRC, the European court is currently adjudicating many GDPR enforcement cases opened around 2020.
It’s plausible AI enforcement will be different, but my current (weakly held) prior is that the European courts would take so long to deliver any judgement or do anything as to be approximately irrelevant here.
I almost missed this! I think this would be a good top-level post, with transcript included.
I am sure many of them. We get a lot of random LLM slop, some of which seems downstream of agents doing things at least somewhat autonomously. I haven’t seen any agent try persistently or do anything clever.
People don’t like probabilities. This isn’t how they think natively. That sucks, but maybe we need to meet them where they are at, and talk in their native epistemic representations. And definitely don’t say “I’d bet on x-risk if only there was a financial instrument for this”. Read Hanson on how people feel about profit.
Man, I find it super weird for this to be a lesson to take away from recent events. The probabilities are the thing that I’ve seen quoted dozens of times and that seem to have caused everyone to freak out (rightfully so). Like, there were like 5-10 big media interviews were people were like “holy shit, 10%, what, that is obviously far far far too high to be acceptable”.
It has been updated, did I miss anything? We now have proper blockquotes and horizontal rules and the jazz.
I highly doubt LW is supportive on net, mostly just indifferent. LW is generally less concerned with vaguely adjacent organizations than the EA Forum, but the people I know think this is quite bad/dumb.
I think the AI Risk Mitigation Fund will do pretty comparably good work!
Thank you!