I'm writing a book about epistemology. It's about The Problem of the Criterion, why it's important, and what it has to tell us about how we approach knowing the truth.
I've also written a lot about AI safety. Some of the more interesting stuff can be found at the site of my currently-dormant AI safety org, PAISRI.
This morning, I put everything I have written into its context window and then gave it this prompt
I use Claude to help me with various writing tasks. Until 4.5, if I gave it a prompt like "write about x in my style" when giving it maybe 50k words in my style, it'd produce something only vaguely in the direction of how I write. Now, 4.5 is able to more reliably produce text that's closer to my idiolect and regresses to the mean less, though not quite enough that I'd be happy passing of Claude's output as my own.
(The actual way I usually use it is to help with editing by helping me get over blocks in what to say next. So I talk to it about what I'm trying to say, get it to draft text, and then use that as a starting point that I completely rewrite. I don't claim to have a great flow here, but it at least lets me iterate rather than have to wander off and wait to get an idea about how to get through whatever is not working about the text.)
Interesting post!
What it feels like for me to understand something is that I have a mental model of its gears and can trace through their interactions. This is usually a mental picture with a bunch of details I can explore and trace through. Usually it feels like a circuit diagram (a biased bit of my ontology that's the result of my education) and I'm literally tracing the flow of electrons through the circuit to see what signal comes out the end or what actuators get activated. This is true even of my model of things like mathematics, biology, and other people.
My models are also full of black boxes, though, where signals go in and noisy signals come out. When I don't understand something, it's a black box. As I study the signals, I slowly work out a theory of what's inside the black box. Sometimes I get so far as to have a nice circuit drawn up where the black box was. Other time I just have a note pinned on the box with my findings. Over time I try to convert black boxes to inspectable circuits.
What does "control" mean here other than "determine the behavior of", and a person cannot fully determine the behavior of another person. To say we can control someone is to be confused about how the world works. We can only exert some influence, and we have fairly limited capacities for modeling the effect of that influence (yes, we are better at modeling that influence than other animals are, but still quite limited).
It sounds like what you are actually saying, translated into stuff that I care about, is that "you are incorrect that a large number of people will predictably respond negatively to this." Or, do you think that's correct, but, still incorrect to care?
I'm not saying large numbers of people will respond negatively or not. For example, sometimes I need to send a wall of text when having a technical discussion because many words are the only way to be precise.
I'm instead saying the mind that cooks up the wall of text strategy is the mind that believes it can exert more exacting control of another person's thoughts than is generally possible. Pithily, I would say you can't control another person. Less pithily, I'd say the level of influence you have to direct the behavior of another person via a wall of text is fuzzy enough that you are tricking yourself into being trapped in a local maxima by doing so (and this usually happens by projecting one's perceived control of one's own thoughts into a believe that by controlling the counterfactual model in your head you can control the world) and you should abandon this strategy for one that let's you just interact with the world and see what happens rather than blowing past the limits of the precision of your model.
No, it's totally worth it to be polite or otherwise execute a communication strategy that in expectation satisfies your concerns. There's no reason that needs to include what I might uncharitably call a neurotic fixation on how another person is going to perceive you, or what I would more charitably describe as an attempt to control how others think and feel about you.
Is there any sense in which anyone controls anything given this framework?
Indeed, no, which is the point. The very idea of "control" is the result of making confused metaphysical assumptions about causality, but I admit that talking as if we have "control" of our own actions is usually useful, as in we only have control over what we do, not what others do.
I think you would be more clear here if instead of saying "you don't have control", you said whatever thing you think goes better if you're orienting the way you are imagining orienting.
My original comment is reacting to what I view as the framing of this post, which I read as supposing a belief that maybe we can make a conversation go right if we just say the right things. Now it might be that if you say certain things another person will react in the way that you want, but the level of influence here is not strong enough that gaming it out in great detail is likely to be useful. We even see this in Ruby's story, where he ties himself up in knots only to get back an "lol i do the same thing" message that defuses the tension.
Instead we can say things, see what happens, and become better over time at saying things that result in the effects that we want. Getting caught up in the meta-message is letting reason run ahead of evidence. You might read my original comment as advice to be brave enough to just say things and learn from the saying and not worry too much about making mistakes, as most people most of the time are unlikely to make truly catastrophic mistakes.
Obviously how you present yourself affects how people perceive you?
Yes, of course.
I can totally control how people perceive me by choosing, say, whether or not to yell racial slurs at them for no reason.
This is factually false. You have zero control over how someone perceives you, no matter what you do. You instead have a model that predicts how they will respond and you can use that model to review a set of predicted outcomes and choose the actions that predict the best outcomes, but that's substantially different from controlling how someone else perceives you.
You have to actually say something to find out how they will respond, and even then you will only get a noisy signal about their actual experiences.
There's also a meta-meta-message behind the messages to avoid sending the meta-message, and it's "I can control how other people experience me". Let go of that, and you'll be free to just send messages, no matter what the meta-message is.
Only once completely dead can one begin to fully live!
Sure, dropping it seems fine. I'm a bit hesitant to address your points too directly because I anticipate the conversation would require something like explaining our entire worldviews and me trying to convince you of mine, so I've been doing something more like just explaining my worldview as I see it as relevant to the points here in the hopes that it gets you to see what I'm pointing at, but if that's not working no reason to continue since I'm not excited right now about making time to explain the whole thing.