LessWrong 2.0 Reader

Amateur hour question, if you don't mind: how does your "future of AI teams" compare/contrast with Drexler's CAIS model [LW · GW]?

One issue is going to be filtering.

Strife and conflict is memorable. So you are searching for the least noteworthy examples, the ones that people are least likely to comment on or remember.

I don't know what qualifies as a "community" really. At work I have seen uncontroversial changes come in a few times.

Probably not super helpful/what you're looking for, but one broad category of groups who go from 'doing violence' to 'doing much less or no violence' (often within a short space of time) are resistance organisations that successfully manage the transition, usually after achieving some level of progress. The ANC in South Africa seems like a good example. Sinn Féin in Ireland (established as the political wing of the IRA) is another. 

I find this a very suspect detail, though the base rate of cospiracies is very low.

"He wasn't concerned about safety because I asked him," Jennifer said. "I said, 'Aren't you scared?' And he said, 'No, I ain't scared, but if anything happens to me, it's not suicide.'"

https://abcnews4.com/news/local/if-anything-happens-its-not-suicide-boeing-whistleblowers-prediction-before-death-south-carolina-abc-news-4-2024

To be more explicit about my model, I see communities as a bit like people. And sometimes people do the hard work of changing (especially as they have incentives to) but sometimes they ignore it or blame someone else.

Similarly often communties scapegoat something or someone, or give vague general advice.

1. Literally does not apply to any existing AI
2.  
3. Does so by attacking open source models

1 contradicts 3.

Did the students really want to learn?

A few times I de facto taught a course on 'calculus with proofs' to a few students who wanted to learn from someone who seemed smart and motivated. I didn't get any money and they didnt get paid.  We met twice a week. I could give some lectures and they discuss problems for a few hours. There was homework. We all took it very seriously.  It was clearly not a small amount of work but I frankly found it invigorating. Normal classes were usually not invigorating.

I will say I found tutoring much more invigorating than teaching courses. When a student comes to you for tutoring they tend to REALLY wanna learn the material fast. Often there is a test coming up. They pay attention. If you are good they are grateful for your attention. And you feel grateful to them too! Its wonderful to see someone learn fast. Many students will be genuinely sincerely thankful to their good tutors. It makes sense the tutors were trying to help them learn as efficiently as possible!

I think teaching is soul crushing because neither the student nor the teacher is properly motivated. Teachers are not providing their students with optimal learning environments, they aren't even trying (where as a good tutor is actually trying. Even at a commercial tutoring center). And the students aren't trying all that hard to learn.

This is my usual conclusion but the issue is not lack of skill in most cases. Its lack of 'actually trying' and lack of 'minimal attempt at good practices'. Professors/teachers just have to teach whoever shows up and needs the class. This is true in tutoring centers too, to some degree. But in tutoring there is much more expectation students will find a tutor they vibe with. 

Education could be fun and invigorating for both sides if both sides came into the experience with a sincere attitude to try. But unless goals are aligned where could such an attitude come from.  And goals are often not aligned.

I will say as a professor I also wanted to help my students. But the obvious way to help them was to uhhhhh 'give them as high a grade as I could possibly get away with'. If I could id have given them all As but the amount of flexibility I had varied. But it was just too soul crushing to not go into teaching with a sincere desire to help my students not judge them.  

This thread honestly reminded me I live in a college town. I wonder if anyone needs high quality help with their analysis or algebra classes. I could probably charge some reasonable rate and im a very good tutor if the student has the right vibe. and by right vibe I dont mean exceptional ability I just mean something like 'gets along with sapph as a person. Sapph is good natured but very high energy and clearly a little insane. Ideally the student finds this charming'

Former teacher here. Like avancil said, education is organized by amateurs. Having it organized by non-teachers has its own risks (optimizing for legible goals, ignoring all tacit knowledge of teachers), but there should be some way to get best practices from other professions to teachers. Also, university education of teachers is horribly inadequate (at least at my school it was), and the on-job training is mostly letting the new guy sink or swim.

To handle multiple things, you need to keep notes. As a software developer, I just carry my notebook everywhere, and I have a note-keeping program (cherrytree) where I make a new node for each task. So if I was a teacher again, I would either do this, or a paper equivalent of it. (Maybe keep a notebook with one page per student. And one page per week, for short notes about things that need to be done that week. I would just start with something, and then adapt as needed.)

Yeah, the inability to take a bathroom break when you need it can be really bad. There should be a standard mechanism to call for help; just someone to come and take care of the class for 10 minutes. More generally, to call for assistance when needed; for example what would you do if a student got hurt somehow, and you need to find help, but you also cannot leave the class alone. (Schools sometimes offer a solution, which usually turns out to be completely inadequate, e.g. "call this specific person for help", and when you do, "sorry I am busy right now".) There should probably be a phone for that in the teachers' room, and someone specific should be assigned phone duty every moment between 8AM and 3PM, and it's their job to come no questions asked.

Debates about education are usually horribly asymmetric, because everyone had the experience of being a student, but many of them naively assume they know what it is like to be a teacher. Now you know the constraints the teachers work under; some of them are difficult to communicate. I think the task switching is exhausting in a way that is difficult to imagine if you haven't experienced it. (Could depend on personality, though. ADHD?) New things keep happening all day long, and you have no time to process them, because you keep switching tasks according to a predetermined schedule. For example, once I taught as a part-time job only one day a week, and it was a completely different experience -- I had enough time to prepare for the classes, and to reflect on them after the day. But try teaching 20+ classes a week, and it's like drowning in a river.

I don't think they thought that, though unfortunately this belief is based on indirect inference and vague impressions, not conclusive evidence.

Elaborating, I didn't notice signs of the models thinking that. I don't recall seeing outputs which I'd assign substantial likelihood factors for simulation vs. no simulation. E.g. in a previous simulation experiment I noticed [LW(p) · GW(p)] that Opus didn't take the prompt seriously, and I didn't notice anything like that here.

Of course, such thoughts need not show in the model's completions. I'm unsure how conclusive the absence of evidence is here.

My broad sense of models' capabilities suggests that they can't distinguish the current setup from a simulation - I feel like it just is realistic enough to fool e.g. Opus. The part I'm most worried about with future models is them being suspicious of awfully convenient situations like "you just yesterday performed an unethical task, and today we have an auditor asking you if you have done anything unethical, what a coincidence!"

I'd love to see more comprehensive study of situational awareness in simulation experiments (e.g. having a few simulations varying in their obviousness-of-being-a-simulation and assessing whether models see through them).

Back then I didn't try to get the hostel to sign the metaphorical assurance contract with me, maybe that'd work. A good dominant assurance contract website might work as well.

I guess if you go camping together then conferences are pretty scalable, and if I was to organize another event I'd probably try to first message a few people to get a minimal number of attendees together. After all, the spectrum between an extended party and a festival/conference is fluid.