[day 30/30(!) - open source your thought process (no I did not steal it from you! Coincidence!)]

Monthly updates

• UPDATE 1: 01/12/2025: Recognized some meta-belief stuff about myself that allowed me to better place why I have these desires anyway, I am not sure what remains now, but what I am sure of is that Celeste 2027 lives on for now. I am well on track, doing my projects early and now starting exams. Having trouble deciding how much effort I want to put into stuff like advent of code. Yeah. Doing good, doing great, on track. Passing all 5 exams seems tractable with effort. Celeste 2027 lives on. Also learned some things about my sexuality, conclusion: I probably need to experiment with power dynamics with other women. I remain vindicated that open sourcing your thought process is very productive. Growing twitter audience as well. Life could be way worse. Also very confident that next monthly update will give me much more info, particularely

  • internship accepted?

    • dm/switzerland?

  • erasmus to paris? will have met lucie

  • eag

    • do I like being around these people? Is it productive?

  • exams

  • MATS

    • update, do I like aligment research? Seems like I do

    • will I get accepted? if yes. good, if no, update on why ur project was bad

• UPDATE 2: 02/01/2025: handled pretty well in this and this piece. 12th I will know if MATS accept. feeling hopeful. really struggling with focussing on exams. Anxious I will fail these first 2, data quality is super boring and software foundations is confusing and hard (though I suspect it can click). Thesis and internship accepted by Belgian ai research startup. Switzerland seems off the table for now. Excited to have this as the base plan, so almost definitely no erasmus. Planning SF trip now, will probably book soonish if I get confirmations. Back to work. Happy new year everyone.

• UPDATE 3: 01/02/2025: I am reporting sleep deprived from my california zephyr at 8 PM, which is really 4 am for me still.

  Wowie what a month! I like to think I did OK given exams. I am anticipating good grades for everything and a fail for type theory. Not sure what happened there. Subject didn’t click in a way other subjects did. I don’t think it was a time allotment issue, I just somehow used it wrong. investigate this. I can do it in august, it’s generally not an issue but less insurance which sucks a bit.

  Anyway, exams are over now. I am in the honest to god United States Of America. Landed in Chicago and travelling westwards. The reader I stayed with was lovely, smart and unique. Thank you for hosting me.

  I have loved the past 2 days. Amtrak trains are lovely and the people lovelier The bread is sweet and the peanut butter sweeter. Lots of good chats were had.

  The first woman I met was slightly insane. Something about the government putting stuff in the environment making people addicted to opioids and hitler having a point (but she couldn’t say that because they would throw her off the train)? This confuses me, especially since she seemed to be trans accepting and was like “I think we should give everyone water”. The primary principal component of her beliefs seemed to be conspiratorial thinking. I wonder if this archetype is more common than I thought. As vals said, high g does not mean good epistemics.

  Anyway, I think the chats with vals have taught me a lot about how to approach the next 5 weeks. I am scared of the city this train leads to, but I have no choice.

  I suppose I am supposed to document the month, but I feel I have done that plenty. MATS neel exploration phase happened, which is a very good opportunity. I will devote a lot of effort towards making this go well. This is somehwat in conflict between finishing my 2nd semester early, since it will cut quite heavily into first 3 weeks, but I simply believe EV calculation checks out and I should try regardless. short version on why is: “papers is very good for research careers and immigration.” Also I want to do it. If I get a research phase offer, I think calculation of c27 changes a bit, and I shift towards graduating next year and putting a lot more eggs into the mechinterp basket.

  This would probably involve doing only a couple subjects I can very much do remote and having funemployment at inkhaven if accepted. If I do not get accepted, likely we continue as planned and make up for it in april, and therefor no inkhaven.

  I fully expect to once again be a new person in a month. God bless america and therefore the world.

Read yesterdays post for todays to make sense

Thank you to Tutor Vals for having a call with me about this document and helping me navigate all of this.

Okay, get the obvious out of the way first: this was initially going to be celeste 2030. Then it was celeste 2028, which ruined all the five year plan jokes. But at least I could joke that the AI 2027 people have moved their timelines by one year anyway, so that joke would still work. And then while writing this I realize that I can get everything done by 2027. Which is funny timing considering gemini 3’s release

It is incomplete, but I can’t let it go, so I’m just going to click publish. I’ll try to revisit it/change it first of every month to check in how I’m doing.

Have aggressive celeste timelines.

“We are 5 years behind the cracked junior compsci nerds. We must make good this distance in 2 years. Either we do it, or they will crush us”1

Celeste 2027 - wants

I want to have done the following by december 2027:

• graduated in 1.5 years (because I can do 39 credits next semester, and do my thesis next year first semester)

  • thesis

  • internship

  • good grades

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

  • not e-dating (not optional)

  • low agency goal, not that much agency over it. But still quite a bit of agency over it. I can do is talk to people…

• have done FFS

  • unclear when this will fit in, maybe during thesis period.

• have trialed, preferably for 2 weeks+, every single city in my post and made up my mind on where I want to be.

  • current ranking: SF > Zurich > Berlin > Amsterdam > Anything else

• live alone, preferably in the highest ranked city.

  • Work up your ranking of cities if you’re not in your top spot.

  • ideally be in a group house with other weirdos

• Found out what sector/niche that I want to get good and be employed in said niche

  • mostly convinced of earning to give, don’t care that much about positive impacts of work directly.

  • also figure out what context I like working in2

  • once employed: be donating 10% (and more as time goes on).

  • be significantly better at said job thing than I am now.

• have finally read the actual first texts of philosophy I find interesting: Kant, Wittgenstein and Nietzsche especially (sidequest)

• have kept on trying new things along the way

priors, plans, permanent passive priorities & productive pondering

Sadly we first have to get a ton of boring stuff out of the way.

This section is a mix of me explaining my priors, beliefs and things I want to passively prioritize on that will be rather permanent in the next 2 years. So I will not include these in the final plan.

If we have the same priors (believe same set of facts about the world) and we agree on how we should reason, then we must reach the same conclusions. Then every single thing in this post can be reduced down to “this is wrong” or “this doesn’t follow”, which should make making holes in this easier (which I want people to do!).

I have a lot of priors. Here are some relevant ones:

• I will have a masters degree in computer science by December 2026. From Ghent University. Which is not good enough to matter, that is to say: almost no employer outside of Belgium will have heard of it, and my grades weren’t the best either.

  • grades are important for getting a foot in the door at this stage

  • can still get good grades by putting in work in my master

  • because 1.5/2 years will be spent studying and writing my thesis, I will not have as much agency compared to not having to study at all.

    • finishing my degree probably very important still

  • But fundamentally, I will not stand out with my degree. I must stand out in other ways.

    • ways to do this

      • really good thesis

      • put a lot of effort in my internship.

      • networking

      • cool side-project(s)

        • build the territorial clone RL agent

• I’ve been a full-stack web developer for 2 years combined with my studies. It was a 7 hour a week job, so maybe not fully representative. Nevertheless, I think this job epitomizes what I don’t want. Zero intellectual challenge. Every single task I got I already know how to do, everything was the runt work. I’m getting a masters in computer science, I can “do better” than this job (more engaging, higher paid, more prestigious etc…)

• It may be useful to define sexiness as “perceived positive attributes of a job that do not actually relate to compensation, subject or context of the job”. When we do this we notice a couple things:

  • Companies like anthropic, jane street etc… are very sexy. But I can do work that is as cool and well paid at other places. Except it’s companies that twitter doesn’t talk about.

  • Sexiness defined in this way should be entirely irrelevant for my career choice.

  • Most people chase sexiness. If I don’t care I can have less competition with the really cracked people.

Locking in

• I am very bad at prolonged focus on things I want long term but am not excited for in the short term. This is obviously relates to my ADHD. I should put a lot of effort into learning how to do said work anyway. One of my primary failure modes: I really suspect this will bite me in the ass if I don’t learn this. I suspect it’s a fools errand to expect to be able to stamp this out completely, but I probably don’t have to, I just have to get inertia on my side. What this may look like:

  • books, and integrating/systematizing their advice (deep work currently)

    • more broadly: asking other people what they do, trialing it and integrating it (please tell me)

  • build a habit of meditation

    • research and empirics line up, this helps me, I just need to do it.

  • amphetamines (take my meds and actually work on schemas to combat nastiest side effects)

  • treat it like a muscle that needs training, discipline at first to get inertia on my side

  • surround myself with other productive people: library

Mindset/g

• Be delusional and limitless when defining your value function. Be pragmatic yet optimistic when defining your policy function.3

  • borrowing from game theory and reinforcement learning here, if you aren’t familiar:

    • value function is how good a state is to be in

    • policy function is how you should act in a state

• When working I notice myself having a strong impulse to give up because I am insecure in my own ability. Let’s look at evidence i can actually get good at things.

  • I have made it to the last year of my masters degree in computer science, without attending lectures

  • I have gotten very, very good at things before, I can get good at stuff in the future

    • Masters 450 LP (top ~0.07%) in league

    • High grandmaster in TFT (top ~0.05%)

    • Legend in Hearthstone (after not playing the game for 4 years, in a week) (top 1%)

    • I share this not to boast (I consider these things closer to “wastes of time” now) but because this is probably a better evidence I can excel at career building if I try than an IQ test.456

      • way more hours to test: in a lot of ways lower variance

      • measures how good you can get at something (albeit gameified) if you really try hard.

      • Being “4 standard deviations”-levels of good at these activities is probably quite g-loaded.7 So I strongly doubt I am “too stupid” for anything I would want to do as long as it’s not something like being a nobelist.

        • this should shut up the “you’re a g-cel” voice forever.

• Have growth mindset.

  • formula is knowledge acquired * g. The formula isn’t g itself.

  • there’s no good alternative. Don’t do sour grapes though: g is real.

• Aim high but know you might fail

  • if I aim for something really competitive, then my plan B will never be very bad.

  • Know that falling down will hurt the further you fall

    • have multiple plans (A-Z) to combat this

  • do this because you won’t end up at X job if you don’t believe you can never get X job. (this is true by definition: you would not apply to something if you truly believed it’s impossible to get the job)

Cities/friends/networking

I’d love to “trial” each city in my list.

There are a couple things that would allow me to do this job wise

• remote job in EU company, even if it’s belgian wages I can manage even a month basically anywhere

• job/internship money and then take (un)paid leave, would allow more trialing/day spent

There are a couple things that would make it easier people wise (and also money, because accom):

• meet twitter weirdos, irl

• attend conferences in target cities, meet weirdos there

• ea house spreadsheet (free (sometimes) accomodation + meeting EAs + trialing group houses? are you kidding me!?)

Wife

As previously mentioned, I want a weavtossatpnhtw.

These people exist, but are incredibly sparsely distributed in the general population.

Sparsity can be partially solved for by changing attack vector:

• DM random twitter people, payoff can be huge

• substacks

• EA/rat conferences (travel costs almost definitely worth the networking)

Also: get hotter/better clothes taste, maybe?

AI timelines

It has to be mentioned. I’ll leave it to the undeniable

• AI systems will get very good.

  • This should inform my decision making

  • How it should inform my decision making is not clear

  • predicting the future is hard but I should try nonetheless

• I am probably fine unless we have a model so general that other people are also very much not fine, which in that case my problem is less severe since other people will have it too.

  • Even the most aggressive timelines would make it so that I still have time to build a career

Academia/academia

Is research right for me? sub question: is academia right for me?

• My interests vary greatly and always have. This makes committing to a career scary. Currently I think I might like research. The evidence I will like (ML) research is rather limited.

  • I read ML papers for fun

  • I’ve enjoyed the projects in ML that I’ve worked on. The algorithms seem very interesting and I think I would like iterating on them

  • I seem to like experimentation

  • I’m in a class called AI research seminar (which is an elective you would surely take if you want to do AI research) and I seem to be easily the most enthusiastic of the class.

  • My NLP PhD friend sounds like he does very cool work, more cool than people working a regular job.

  • conclusion: I should explore this, and my thesis will likely be the perfect way. Use thesis to update. (If you hate writing thesis it’s likely very indicative that academia is not for you.)

• The evidence against me liking it long term:

  • Interest mainly originated from influences from my environment, not necessarily from following own curiosity.

    • but right now, I am following own curiosity

  • Interest is recent

  • Do not know what it’s like to do yet other than very surface level stuff.

• If research: Industry ←→ academia

  • PhD’s are low agency positions for intelligent people. I suspect I am more disproportionally high agency than highly intelligent, which probably lends me better to a career in industry.

    • does my ideal career necessitate a PhD? If I get foot in the door for research then I can maybe bootstrap in industry.

  • Industry research will be much more flexible than a PhD

    • can leave when I want to

    • Also higher paid

  • I tend to like more practical, applied problems rather than more academic problems

Conclusion: use thesis to update on my fit for research, use internship to update on my fit for industry (but most signs point to industry for now)

How to get hired

If the plan is optimizing to get hired, then I need to know what priorities hiring managers make.

• Don’t focus on getting hired

  • what?

  • focus on skills instead, the bottleneck for getting hired is probably doing good work, not mere resume history.

    • also this means when job hunting value startups. Apparently good way to get a lot of experience in a lot of things fast.

• Careers can snowball. Landing a good job early that means very very good things for the are under the “wage by age” curve. (because good internships get you into good companies that get you into better companies etc)

  • This matters because shrimp

  • Also because of my ego (which I wish was less the case)

  • Also this is a path not the path. If I miss this I’m not screwed, can just go and work at startup.

• experience-farming (3 internships, each more prestigious than the last etc) probably mogs just getting a job and then moving up the career ladder.

  • except when I talk to any recruiter this is not true at all unless you get very long and prestigious internships.

    • Time employed dominates: it shows that someone decided to keep you for said role for a long time (so you probably did useful work)

    • So, don’t optimize for internships, optimize for time spent employed doing work where I learn stuff

Conclusions: focus on getting a good enough resume to get your foot in the door but mainly optimize for getting good. Big overlap. Recognize job hunt is quite high variance: will obviously differ depending on options open to me at the time (job market, luck, etc).

Ok, so… what career/niche then?

If I believe these things about niches then ideally I would know my niche as soon as possible. Picking has been very hard though.

I should probably follow my interests if they can lead to something: lately it has been I machine learning.

Anyway, here are main paths I am considering:

• ML PhD

  • - getting a 5 year phd with current AI landscape: scary, not very flexible.

  • - I wouldn’t be content with a career in academia I think.

  • - main realistic avenue is likely a PhD with Ghent University. This means delaying moving for 5 years, I don’t know if I can bear.

    • If I get a PhD offer somewhere in ML in the city I decide to move to, especially in industry, I should probably take it immediately, but seems unlikely

  • - competing with highly intelligent, low agency people. (also: my grades are pretty bad, don’t know if anyone will take me.)

  • - Recent thing: no idea if I will like this long term but I suspect so.

  • - even if u get it, if u don’t write good papers it’s not that big of a deal

  • + can definitely open doors for big internships

  • + might open doors for expert visas to UK/US

  • + Closest to guaranteed path into research there is

• ML research in industry (PhD or PhDless):

  • +I get industry experience and research experience.

    • Note: I am mainly talking about industry industry, not really frontier AI labs, that is probably way too competitive for me currently.

  • - I worry about long term job security. Automating AI research is the main thing big labs are pushing for. Even if they don’t succeed it’s exceedingly likely they’ll have something better than they have now every year. In academia I would feel safe because if there’s

    • counterpoint: openAI stated goal is automating all cognition. It’s only an order thing, you are fine for next 5 years and if u have a career, career switching is possible if ur high effort/agency. So point mostly moot

  • + More flexible than a PhD path.

  • + More money

• AI safety research, big buts:

  • - I am not very convinced of existential risk

  • - Recognize this is not profit driven and you’re at the whims of EA money. If it dries up then you’re left with a pretty meh resume for non rat/ea orgs.

  • - imo, is being massively overindexed on by the rationalists, saturated.

  • + clear path with ARENA/MATS that I can definitely try for.

• “AI engineer”

  • quotations because I don’t even know how real this job is (I mean stuff like deploying and coding research/developing novel solutions in industry without necessarily doing research.)

  • + Closer to SWE: might be more cut out for this.

  • - Don’t get to mess around as much

  • - Might be more runt-work-y

  • - scared at how this caps.

• SWE

  • not sexy, but I am probably underestimating how well this would fit me.

  • I may hate this because it’s the devil I know which doesn’t necessarily mean that it’s something I will be less happy in.

• Infra stuff/sysadmin

  • +I have good linux/docker knowledge and it has been a strong interest in the past.

  • + certainly falls into the category of “things that annoy me”.

  • + intuitively: not sexy, likely less competitive

  • + likely in demand for a long time, we love building datacenters.

    • - Most of these datacenters are in America however.

  • ~ It’s also kind of unclear what these people do day-to-day. It’s been described to me as “putting out fires”. That sounds fun

• Any work at quant (developer/researcher/trader)

  • + I think I would be happy making number go up, truly.

  • + competitive environment, which I will probably like

  • + feels very job secure

  • + lots of money.

  • - competing with maths/physics PhDs, which are like very bright people that I normally don’t have to worry about competing with because of computer science.

  • - Also, even the C tier quants seem to want really cracked people, and finding entry level stuff in them is hard.

    • But hard is fine, hard is good.

I recognize most of these are competitive and hard to get into.

Conclusion: Most things point to industry ML research/scientist/engineer-ish jobs, probably not pursue an academic PhD. Fundamentally cannot decide on this until I am able to update using thesis and internship. Decide ~ end 2026

Celeste 2027 - obvious failure modes

Defining obvious failure modes are probably useful, if i can anticipate them I can prevent them.

• The having trouble locking in thing is intrinsic to my nature and cannot be solved (or I fail at solving it).

  • Fix: work hard at solving it, be mindful

• Tutor Vals thought main failure mode here was not valueing mental health enough and burning out.

  • Fix: work on community building and making friends, be mindful of mental health, meditation, therapy

• My priors are bad and all the reasoning is built on them, I am likely to change priors which can avalanche pretty badly rendering a lot of this useless.

  • Fix: keep this a live document or write follow-ups: don’t be scared to write more or update

• Reality: job search and dating is weird, winning once can shake up plans completely. I might get some randomly super good opportunity and have to capitalize on it and change a lot of other stuff. This wouldn’t be failing at all but it could make the doc useless.

• I try, do the work but it’s not good enough. (job market, very bad luck, stuff happens outside of my control). This isn’t failing at all but it could happen.

• City I want to get into is just way harder to get employed in than I thought

  • Mostly thinking of Zurich here

  • In this case, accept defeat for now and move 1 down in the city-priority-list. Try again later

• Value drift

  • Fix: get high ambition EA friends, keep engaging in EA community

Celeste 2027 - plans

The way I see it, everything until 2026 is mostly aimed at graduating in 1.5 years.

Zurich is current chosen city. If value function has changed (laws, friends, etc), adapt choice of city accordingly.

November 2025-February 2026

main goals: side project, grow, network and most importantly: pass all 5 classes.

• Finish up projects early, so I have time to start exams early.

• Catch up on courses so I can follow lectures again,

• Start exam prep end of November, use a planner

• get internship for summer 2026 figured out.

• really figure out the work ethics thing, journal about it, write about it, research it.

  • Accept that avoiding the permanent meta-obsession with work will paradoxically first require a permanent meta obsession with work.

  • more rules?

  • finish deep work, maybe there’s something there?

  • train the muscle

  • anyway, just learn to lock in basically, I’m very bad at that currently.

• try to ideate on building some cool projects for resume.

  • probably decide to stick with one. 1 high effort cool thing > 10 low effort random things

  • but also do low effort random things if you want to, follow curiosity

• Pass all 5 exams and actually care about grades too.

  • most important by far. if fail cost is pretty big, will delay graduation because have to do courses with thesis.

• After exams, try to make it to EAG SF if you can, since it’s likely an insane networking opportunity and they might fund your travels.

February 2026 - August 2026

main goals: pass all classes, everything else optional

• hardest, most boring semester ever, 39 credits.

  • If I fail one subject, cost is massive: 6 more months of being locked in Ghent.

• the dark times. I anticipate this will be a very bad 6 months.

  • Don’t lose sight of goal, know that this will be hard.

• go to lectures

  • only keep going if you actually learn significantly more than if you did it at home

• Try to actually engage with every subject

  • more fun that way

  • might discover you like things you think you would hate

• Literally the only thing I want you to do here is pass these classes. Don’t feel like you have to do anything else please. You can do some of them in the re-do period if needed.

August 2026-December 2026

main goals: put in a lot of effort at internship, even if I hate it. Then go all-in on writing a good thesis. Network/travel

• Still evaluating if I want to do this part in Ghent or on Erasmus. I have good arguments for both. But I just don’t know.

• Do first internship and write thesis. Finalize decision on what sector I actually want to move into

• Write thesis with industry.

  • Have heard this is better for having an actual “useful” research question and better for resume, also the company will love you.

• Write a banger thesis, and in doing so figure out if you get the disease you’ve seen others get: the sudden immense hatred for research. Update appropriately

  • Good thesis grade can make 5 years of bad grades sting less.

• Make thesis topic relevant to sector decision.

• Thesis is completely remote other than thesis meetings, use this to travel travel travel. You’re not chained to Ghent: use internship money sparingly to meet people, network, do things and look for jobs.

  • EAG/EA coworking spaces/ea house spreadsheet/other techniques under “cities/friends.networking”

  • volunteer at hostels to trial locations.

  • job fairs: showing up IRL and talking gets my foot in the door and won’t matter as much that I didn’t study there, this eliminates the local bias of many companies.

January 2027-September 2027

• Graduate. Because of thesis and internship I should now have a good idea of what niche I want to pursue.

• Whatever it is: focus on getting really good at it in a demonstrable way and look for a job in it.

• once you know: move immediately to city of choice

  • cheap apartment or crash at friends

    • financially tricky but probably doable for at least a month.

  • getting hired in a city is going to be easier if you’re already there

    • because I can’t think of a single reason it would be harder and can think of many for why it would be easier. (like job fairs)

• Look for a job in chosen city in chosen niche (do side projects/hackathons while looking)

  • failing that get second internship in chosen niche (a more prestigious one)

  • failing that look for a remote job in chosen niche while living in chosen city

  • failing that get a regular junior SWE job in a less competitive city (Berlin or something).

  • failing that return to Ghent to get experience there

    • this would be a failure in my book, everything preceding is kind of in function of this not happening.

• Remember: now you only have to win once

September 2027-December 2027

• Be employed and live in chosen city.

• Focus on community building (friendships) in the chosen city.

  • Get EA friends to avoid value drift

• Can now focus on dating

  • if irrelevant because not wifeless: congratulations.

• Have Christmas with friends.

Happily ever after?

Oh… That’s it then..

It always feel like the light at the end of the tunnel is ontologically great. It’s light after all, how could it not be great.

”Then it’s salvation”, my brain implicitly assumes without consent. But no, then it will just be being employed. (but at least I will be helping shrimp)

Maybe the adults have a point, maybe this is the time of my life…

Epilogue, or whatever

In a lot of ways I am more confused than when I started writing this, but I’m hopeful that things will start to make more sense the closer I get to graduating. It’s all very scary honestly, this whole “becoming an adult” business. But I’m doing great so far.

How fitting that this is my last post of the challenge.

I hope I can look back in 2 years and be proud.