do you want to make it or do you just want to be held?

reflections for 2026

This one just kept growing.¹ She’s being vulnerable on main again. Inspired by this lovely post.

Inside of you there are 2 Madelines

B: It warrants questioning. You just. You know that this is something you’re capable of doing. It wouldn’t be the first time. You convinced yourself you were straight. And the best narrative we’ve found for why you did that (barring social acceptance) is wanting to be held.

C: but I want to go to SF. I want to do AI research. I deeply want all this. I thought we had established this was irreducible? That even if you stripped everything away this is what I would want to do?

B: irreducible does not mean unconfounded². What is it that drives you to SF? What drives you to AI research?

C: Well I mean, it’s where the things happen. AI is probably going to be really transformative (we’re making computers that think!), and… I don’t want to be on the sidelines.

B: are you sure that’s it?

C: Okay. It’s the people too. SF seems to be peak weirdo. The people there seem so cool, I frankly feel inferior.

B: Why inferior?

C: Well because they’ve all “made it” you know. To be in SF means you must have a job that warrants it. Those jobs tend to be really cool.

B: That’s not true. There’s window cleaners in SF.

C: Okay, well, to be in tech, generally.

B: Oh, and why would that make you inferior?

C: Because somewhere along the way I internalized that successful people are better people? Unironically? Like. It’s just cool to be part of something at the frontier? I look up to it?

B: Right. This sounds mimetic, and a bit stupid, do what you love.

C: Maybe it is you know. I don’t really need or want to figure that out. I just know I want it.

B: There’s more to the idea of SF anyway isn’t there? What’s the core emotion associated with it?

C: Yeah… I think it’s escape. From the mundane and the loneliness. It’s always been fucking loneliness. It’s why I stayed in that relationship for as long as I did.

B: Why wouldn’t you be lonely in SF?

C: Well, surely finding a wife in SF can’t be that hard.

B: Then why hasn’t she done it?

C: Well I don’t know! She has weird standards! She is not attracted to autism, bad taste if you ask me.

B: Hmm. What if you make it to SF, and you get your dream job, but you have to be single forever. You’re not allowed to fuck or cuddle anyone. How does that make you feel?

C: first of all, how the fuck am I supposed to even imagine that! I haven’t even been to SF! But yeah! That thought does make SF feel a lot less appealing!

B: What’s going on there you think?

C: I don’t know! I can’t fucking find love here. dating apps blow. My people seem to only exist online or at these weird conferences? And SF seems to be full of them? So I guess I… God, is this really just about finding a wife again?

B: Well why can’t you find love here?

C: I don’t think there’s an issue you know. People would and can love me. I just… I haven’t loved anyone back like that since him. And the SF people remind me of him, I guess? Not really. But they have that je ne sais quoi. That ADHD-opinionated-smart-neurotic-utilitarian-veganism you know. the weavtossnatphtw³… At least, she did

B: You’re describing her again. Then why not just her?

C: Because she didn’t like me the way I liked her damnit! It wasn’t real anyway. You just got obsessed again. And that’s not the first time! Every time I look up to someone like that they’re unavailable.

B: Do you think you’re not enough?

C: I mean. Ugh. What does that even mean? I wouldn’t want to date me! Not anymore! Before all this I think I would. Now I would want myself to get more credentials first.

B: Do you hear yourself??

C: Yes! Argh! I mean. I don’t know! I just think it’s insanely cool to work at the frontier, and discover things, and have an actual impact you know. Do something meaningful. And marrying that with my one true love, computers, it’s just… all I ever wanted. I wouldn’t settle for a partner doing less.

B: Do you want to be this person or date this person?

C: It’s always been both⁴, you know this.

B: do you want to make it or do you just want to be held?

C: I don’t know anymore.

C: Ideally both.

B: (you probably can’t get in anyway)

C: whatever. My twenties are for trying.

SF

B: So… do you think visiting SF would help with this?

C: maybe. It wouldn’t hurt.

B: Then go.

C: I will go. I mean, I want to go. I planned it, 2 weeks in february. I could take the zephyr too. 550 euros total. (excl food and waymos). I could probably afford it! If vivarium helps, probably even less.

B: and what are you hoping to get out of that?

C: Farm interestingness? Really I just need to see for myself. I just… cannot dispell the mythos of SF without actually going.

B: then book the tickets

C: no.

B: why?

C: Remember that one time in Amsterdam where u got a little bit too high and cried about ur ex of 2 years ago?

B: how could I not remember?

C: Well, that’s probably way more scary when there’s an ocean between home and instead of you know, basically being in the same country.

B: Right. But you don’t feel like you have a home anyway

C: I do. But it’s nuanced. I am comfortable here in a lot of ways. Comfortably uncomfortable.

B: noticing nuance and that things are complicated doesn’t mean you should stop thinking about them.

C: that’s true. I’m a bit too light of a feather sometimes. I’m also just scared no one in SF wouldn’t like me.

B: why would no one like you?

C: I don’t know. I guess she scared me you know. She exponential back offed on my ass.

B: But they already like you. A lot of SF people read your blog now. You’ve been at this for less than 3 months and you’ve already found so many people. Scott Alexander reads celeste-land for crying out loud! Why are you overindexing on her, do you think [REDACTED] represents all of that city?

C: Well no, but I don’t really have any other data you know. And SF people are notoriously flakey and slimy and don’t care if you aren’t building something. And it also just feels like the idea of being with⁵ her still controls me. That this is all her still. I need to let go of that

B: But I mean, why do you care? If someone is slimy or flakey, fuck em? It doesn’t matter that this started because of [REDACTED]. It’s embarassing, but who cares. It’s yours now

C: Well okay. I have other hangups too

B: Like?

C: Well. I mean what if it’s just exceedingly boring? like you go there and everyone is working (because they will be) and I don’t get to hang out with people or they are mean or annoying or I don’t fit in at all

B: then that’s data too.

C: And the opportunity cost is big! I could go to London or Paris for a week! People there that I know, that seem more excited to meet me, that are less flaky, it feels much safer.

B: but SF feels more urgent.

C: yeah…

B: You already know what you want. why can’t you just go?

C: I don’t know. Why am I still scared?

B: Maybe the image of the city in my mind is so perfect it can’t possibly survive contact with reality. So you cling onto it for dear life. Falsifying the dream of SF would be falsifying a future. That which can be destroyed by the truth should be. Or something. What do I know?

C: maybe.

B: maybe.

Job

C: You know, I’ll need a job. 80000 hours of something. I might as well like what I do. If uni has taught me anything it’s that diving very deep into a project is deeply fulfilling to me. I have not found limits to this if they exist. I love reading papers. I love exploring the borders of human knowledge. If I have to get up every day and be forced to do something, please let it be research.

B: Then go do research?

C: But everyone in research is so cracked. I didn’t get to go to UC berkeley or MIT. I have to bootstrap all of this. Even worse, I am not close with like any prof in ML, and I kinda can’t get to, because… yeah, the study situation. Academia seems kinda lame. I want industry

B: right. Why is academia lame?

C: Because all the ML secrets are in these private companies you know? Infinite computer. Fucking around. Building something people like, actually use.

B: I thought this was about the borders of human knowledge?

C: Oh I mean, yeah. The borders of human knowledge are being pushed by google you know. But you’re right, it’s ego too. Of course it is ego. But none of that matters if I can’t even get a job in research

B: You think you’re too stupid don’t you.

C: I mean I do… How could I not? Researchers seem very competent.

B: You understand lots of complicated things. Others think you’re very bright. You’re close to finishing your master, and your compsci peers don’t seem very smart do they?

C: But that’s Ghent. No one cares here. Besides, if I understand something, then it isn’t hard anymore.

B: But don’t you see how that’s non falsifiable? That’s how you never get to feel smart.

C: They gave me the damn number in high school. 113. one standard deviation.

B: But IQ tests are often wrong. You had a gap of like, 30 between verbal and performance. That is highly indicative of a just, badly taken test.

C: That’s co-

B: Furthermore it just…. Doesn’t matter? You’ve already proven that you can do the work? The school projects? The research? Like. You are clearly not too stupid to make progress on these things? Work is the bottleneck. The very authors of the papers you build upon compliment your work, that’s not what being too stupid looks like!

C: But I don’t want to be not “too stupid”. I want to be remarkable!

B: There it is! Do you wanna see if you can be Hamming? then go do the work and see. Stop hiding. How has this insecurity around intelligence served you anyway? Surely there’s been some sort of point to it?

C: I don’t know. I can’t think of any. I guess it’s hedging against the embarrassment of being confidently wrong?

B: Besides, don’t you literally already have an internship lined up at an AI research startup next summer?

C: I mean… I do. And it seems pretty cool. But I’m not sure if that’s exactly what I want to do.

B: What is it that you “exactly want to do”?

C: Mechinterp? I just think model architecture is the coolest thing ever. That one modulo arithmetic paper⁶ just… Changed me. We have these models that do incredibly impressive things, and we don’t know how they work? And I can help?! Like, it’s wizard jizz now. Isn’t it crazy that it’s still wizard jizz?

B: so go do mechinterp then

C: But who the fuck will hire me!? I applied to MATS, we’ll see if they get back. But you know. I’m also just not very convinced that AI safety matters. I don’t see LLMs being goal oriented.

B: When is it not fine? probability?

C: I don’t know? 1%?

B: Well, then it probably matters doesn’t it.

C: I know! I know! but like, it’s language models. I don’t think they are going to be dangerous.

B: even if it were 0.01%. If the stakes are extinction, and you could tilt the scales, the EV is always crazy high

C: But it’s just hard to motivate yourself to do a thing when you just viscerally and rationally believe that it’s probably not going to matter. At least not for language models. And that’s all everyone is working on.

B: Then build things that help mechinterp, not just language models?

C: I mean yes! This is why I love latentqa and activation oracles so much, and why I picked them to do research on top of. If we ever get the really scary RLd brain in a vat, the thing that really scares me, as long as it has activations we could be ready, and easily scale these methods!

B: So it seems you don’t not care about AI safety?… You just have… takes about it?

C: But they’re so unfounded! Who the fuck am I to have opinions on AI safety?

B: Isn’t ur entire thing to have opinions, loudly, and have people prove you wrong? Isn’t that what this blog is? Our entire thing?

C: Well yeah! But not in front of neel fucking nanda! That guy has better things to do with his time.

B: Then don’t have them in front of neel fucking nanda, throw them on LessWrong⁷

C: and what if AI safety turns out not to matter at all, and it’s just in the hands of one google employee. Then what?

B: Then what?

C: Well then it was all for nothing! And I will be unhireable! Mechinterp feels like such a fake science anyway. The ML people are developing new algorithms and making all this new stuff happen. The mechinterp people… I don’t know. it feels like alignment and linear probe discussion #999321.

B: Are you in this for x-risk or for your ego? Is it doing something meaningful or the aesthetics of intellectualism? Do you truly want to do mechinterp or does it just feel more achievable than pushing the ML frontier?

C: All of the above!

B: That’s not how EV works anyway. Even if you pursue mechinterp, skills transfer. But sure, let’s say I grant all of it. You might not be fine if there’s an LLM bubble and it pops. What would you want to do if this was the case?

C: I would still want to study ML models for a living.

B: Then go do it. Be yourself? Do the thing? Have the opinions? It’s fine. Just be competent. If you’re competent enough you’ll have a job. Also, it sounds like you might want to do a PhD?…

C: Well, I mean, a PhD has the opposite problem, being quasi locked into something for 5 years, where I have limited freedom of what to study is also very scary, what if AI turns out to be a super big deal?

B: In that case we’re all fucked anyway.

C: yeah, I mean, duh. But it would be sad if that ended with me being in academia in Ghent. I also just… doubt I could get in. I’m not the “““good student””” like some of my friends.

B: You’ll see if you can get in when you get there. Anyway, if we’re talking about the literal end times: who cares where you are? Why does any of it matter?

C: Because academia doesn’t ship things! Especially Belgian academia! What’s the last meaningful thing Ghent University has done for ML? You look at your professor’s google scholars and it just feels sad. Like none of this work matters, at all.

B: Then fuck it! Don’t do a PhD! Just bootstrap doing research!

C: But how do I get competent enough?⁸

B: By putting in the hours

C: But I’m scared to put hours into a thing I don’t fully believe will matter you know.

B: Yeah, I know, I’m you after all. I don’t know what the answer is. Do it scared

C: Yeah… but what if -

B: do it scared.

1. sometimes it’s all over the place and I’m a bit too desperate to draw a coherent narrative with neat conclusions. It probably would’ve worked better with no editing and just pure stream of consciousness

2. I think I’m using these terms correctly

3. (weirdo effective altruist terminally online slightly schizo atypical thinking pattern neurodivergent hot trans wife)

4. obligatory: blanchard bad sometimes. typology is an incomplete story. Despite that it’s helped me make sense of myself. I am attracted to what I want to become, and vice versa.

5. someone like

6. Excellent explainer here

7. Oh how I wish I could translate Teun’s “dan kan je het altijd even op lesswrong knallen of niet” to english.

8. (and won’t I miss mentorship?)