Huh, that’s a potentially significant update for me. Two questions:
1. Can you give me a source for the claim that making the models incapable of deception seems likely to work? I managed to miss that so far.
2. What do you make of Gradual Disempowerment? Seems to imply that even successful technical alignment might lead to doom.
Severin T. Seehrich
Karma: 647
Offer: Team Conflict Counseling for AI Safety Orgs
Good catch! My implicit question was about what ends up on the frontpage, i.e. some mix of version 1 and 3. A friend of mine answered the sociological side of that question to my satisfaction: Many of the most competent people already pivoted to governance/outreach. But they don’t have much use for in-group signalling, so they have quantitatively much less posts on the frontpage than others.
On a more general note, it’s certainly possible that I vastly overestimate how well the median LessWronger will be at presenting the case for halting AI progress to non-rationalists.
After all, I’ve kept up considerable involvement with my normie family and non-rationalist communities over the past years and put a bunch of skill points into bridging the worlds. To the point that by now, I find it easier to navigate leftist than rationalist spaces despite my more gray tribe politics—because I know the local norms from the olden days, and expect leftists to be more fluent at guess culture so I don’t need to verbalize so many things. In addition, I’m unusually agnostic on the more controversial LW pet topics like transhumanism compared to others here.
At the same time, having constructive conversations with normies is a learnable skill. I suspect that many LWers have about as much learned helplessness around that as I had two or three years ago. I admit that it might make sense for super technical people to stay in their lane and just keep building on their existing skill trees. Still, I suspect that for more rationalists than are currently doing it, investing more skill points into being normie-compatible and helping with Control AI-style outreach might be a high-leverage thing to do.
So you think the alignment problem is solvable within the time we appear to have left? I’m very sceptical about that, and that makes me increasingly prone to believe that CEV, at this point in history, genuinely is not a relevant question. Which appears to be a position a number of people in PauseAI hold.
Why does LW not put much more focus on AI governance and outreach?
Makes sense! My current emerging policy is to keep going out, but to avoid being in closed rooms for less than high ev interactions. Plus prioritizing applying to remote over in-person jobs.
Given the large leaps with which AGI and various other existential risks come closer, being extremely covid cautious seems not indicated though. Gotta see the world while it still stands.
A key piece of information I’m missing here is how well-myelinated orca brains are compared to human brains.
A quick Google search (1) suggests that “unmyelinated axon conduction velocities range from about 0.5 to 10 m/s, myelinated axons can conduct at velocities up to 150 m/s.” This seems even more significant in orcas than in humans given their larger brain and body size.
My covid-related beliefs and questions
Interesting arguments going on on the e/acc Twitter side of this debate: https://x.com/khoomeik/status/1799966607583899734
Tend to your clarity, not your confusion
Awesome, congratulations for the start of your networking journey!
Even though it can be really disheartening, remember that failure is an inevitable part of the journey. Remember the Edison quote: “I have not failed. I’ve just found 10,000 ways that won’t work.”
Yep, I’m currently finding the balance between adding enough examples to posts and being sufficiently un-perfectionistic that I post at all.
My current main criterion is something like “Do these people make me feel good, empowered, and give me a sense of community?” I expect that to change over time.
If a simple integer doesn’t work for you, maybe split the two columns into several different categories? If you want to go fancy, weighted factor modelling might be a good tool for that.
Feel free to adapt it however it makes sense for you. :)
It’s all about the difference: If they are the same, leave everything as is. If “want” is higher than “is”, make some intentional decisions to invest into that relationship more. If “want” is lower than “is”, ask yourself wtf is going on there and how to change it.
I actually told the most hippie human on my list (spending months on rainbow gatherings-level hippie) that she’s on it. To my surprise, she felt unambiguously flattered. Seems like the people who know me trust that I can be intentional without being objectifying. :)
The Dunbar Playbook: A CRM system for your friends
Yea, but I don’t remember claiming anywhere that I can cure anybody’s depression, and don’t really intend to ever do that...?
I did not recommend any particular intervention in my post. I just tried to explain some part of my understanding of how new psycho- and social technologies are generated, and what conclusions I draw from that.
If you expect most if not all established therapeutic interventions to not survive the replication crisis—what would you consider sufficient evidence for using or suggesting a certain intervention?
For example, a friend of mine felt blue today and I sent them a video of an animated dancing seal without extensively googling for meta-analyses on the effect of cute seal videos on peoples’ moods beforehand. Would you say I had sufficient evidence to assume that doing so is better than not doing so? Or did I commit epistemic sin in making that decision? This is an honest question, because I don’t yet get your point.
Huh, interesting. We seem to understand different things under the term “policy” there. Going by the titles alone, I’d only have counted lines 13, 18-20 as “yes” and 33 as “maybe”. So 12-15%. 9-13% if I exclude this post, which has been made after observing the ratio.
Still curious what percentage they make up among total submissions, not only under the AI tag.