Clojurians Log v2

Talking about sports and Denmark. We recently sponsored a game in Copenhague.

neat! How come?

I call it funky marketing. I love basketball and sponsoring international games is not that expensive.

I climbed with a rope and carabiners for a stint. Gave it a decent go, but it was really not for me - I didn't get a kick out of it, only scares. But I've always thought that bouldering could be fun.

It helps that there is a (somewhat hard) mattress below you when you inevitably fall down.

I've been a keen boulderer for over 20 years, and had multiple trips to Fontainbleau in France (the spiritual home of bouldering). I really enjoy the problem solving aspects as well as the physical challenge. It's the only sport I've ever really stuck with.

Yeah, it definitely seems like an interesting mix of physicality and planning your moves. Cool to hear that you stuck with it for so long. It's definitely not my last time.

give it time

for us to loose our heads? or for us to turn into machines? or for this morning to turn bad?

perschonally, i hope to retain my head, my homo-sapiens-club membership, and the goodness of my morning!

actually, any of those three will do at moment....

Any of the above 🙂 I'm reminded of the conserved heads from Futurama, and also of some space opera I read where people belonging to a certain tribe go on to live in a world of shared consciousness that exists inside their biological spaceships when they die.

aroouuuuuuhhhh

@teodorlu are you channeling Gandalf with that comment? 😁

perhaps in some of my better moments! I'm afraid I'm still subject to all the faults of human thought 😄

I just mean that it reminded me of that well-known “good morning” quote from The Hobbit 🙂

ah, right! Probably unconscious Tolkien influence, then. Very nice quote. > “Good Morning!" said Bilbo, and he meant it. The sun was shining, and the grass was very green. But Gandalf looked at him from under long bushy eyebrows that stuck out further than the brim of his shady hat. > "What do you mean?" he said. "Do you wish me a good morning, or mean that it is a good morning whether I want it or not; or that you feel good this morning; or that it is a morning to be good on?" > "All of them at once," said Bilbo. "And a very fine morning for a pipe of tobacco out of doors, into the bargain. > https://www.goodreads.com/quotes/14559-good-morning-said-bilbo-and-he-meant-it-the-sun

That seems likely 🙂

No, I thought about last year. As I now work for the Dutch Railways and the railways have a strong union history. But then one of the unions went on strike for something that I thought was a bit excessive.

Yes, but this is very common in Norway. Not sure what the exact numbers are, but I think around 50% of all workers are unionized here.

52,1% of all workers were unionized in 2024.

While you still have a job, you might try to enjoy it while it lasts? ;)

or embrace the inevitable 😉

I do have a few OS projects that I want to see the light of day, but I cannot for the life of me have both a coding job and a coding hobby

well, not in spring and summer

are you using claude at your job?

@teodorlu you were going to say something, I would love your perspective on this ;)

haha, you never miss anything, do you!

yeah, slowly leaning into it. I've used it more extensively for a green field pet project, it performed pretty impressively.

I like to use it to dig deep in compsci topics, not something one typically does at a job.

I just wish I had the financial independence to do such a thing. Jobs does have an oppotunity cost, but something has to pay the bills! And poking at Claude also has an opportunity cost. We could bike around and write. We could help some school teach technology in a non-stupid way. We could attempt to pursue knowledge in a way not tailored towards getting grant money for our institute.

Maybe quitting for a period is reasonable? Take 3 months off, give oneself a bit of time. Sabbaticals are common some places.

yeah the sabbatical is very tempting

or ask for 20% time at your job to fool around

Yeah of course clauding is not something I want to do in my free time when the outdoors calls.

hmm, but first getting some fresh air and movement in, then go inside for a bit of research/learning/curiosity-following. That could be something.

it seems some anthropic people favor doing poetry over doing claude too: https://x.com/MrinankSharma/status/2020881722003583421

I think another fear is that the daily chores are bad for the learning curve, eventually making you lag behind so much that you'll never catch up with the peleton ahead of you and become obsolete way before retirement time... but then again we might run out of energy to run these AI data centers and the mentats will rise again.

Mentat as in "old school" engineers

fvck Claude and his buddies in the 455 ✊🏼

my poetic thought for the day

How's the AI usage landscape at Exoscale, @raymcdermott?

I recently paired with a younger colleague plus Claude on a greenfield thing at $DAYJOB. It was a very interesting experience that has some obvious strengths and weaknesses, which now has me pondering some things about the future of programming, and many of those ponderings are not particularly optimistic when it comes to the culture of programming, software engineering, &c. :/

@jackrusher can you talk more about the strengths and weaknesses and what you were trying to build?

The technology was originally developed in the context of Google Translate, and it makes sense that it works well with programming languages because they are strictly simpler (i.e. context free grammar, restricted vocab, more repetitive) than natural language. So for use cases that involve translating an algorithm that’s in the training data in one programming language to another language/context, it can be quite effective. The greenfield project was exactly that: producing a library in Rust that implements some well known (extremely common in the training data) numerical algorithms in a way that’s easy to compile to various GPU targets, including WASM+WGPU. After writing a fairly detailed spec and iterating on it in “planning mode” to generate a step by step plan, it was able to generate the set of functions that I would have wanted, and it was able to generate two test cases that were more or less the same as what I would have written, both of which passed. We then manually modified the tests in ways that let us verify that it was going what we thought, &c. This is super impressive and saved maybe two weeks of manual (and tedious) typing here, and I can’t say that I missed implementing the dot product for 1000th time in my life. But knowing how it works and the limits of the training data, I also tried to get it to do something that I knew was possible but also was not in the training data. Specifically, I asked it to propose an implementation of an algorithm from differential geometry using 3D Projective Geometric Algebra to simplify the approach. This is something that had occurred to me in the shower, and I felt sure could be done quite elegantly, but a literature search showed that no one else had ever published anything about it. Claude failed for 1.5 hours in a row, constantly trying to sneak in rewritten pieces of the standard linear algebra toolkit (cross product, &c) so that it could express the algorithm as a translation of the well known method that’s heavily represented in the training data, and it told me the whole time (with great confidence) that it was solving the problem I had put in front of it. After that, I wrote the program in ~20 minutes, and it worked exactly as I had expected. So, if you’re doing rote reimplementation work (as almost everyone is), I think it can be an accelerant, but it’s also very much a snapshot of common practice at the moment when it was trained, and it will not be very good for generating actually novel things, nor pushing forward the state of human understanding (while still maybe serving as a useful if unreliable tutor for the understanding of individual humans).

make sense

@jackrusher isn't the bigger problem (modulo the really big problems with AI) that you knew the answers and there are many cases (probably most) where folks don't. Understanding and evaluating outputs won't come for free.

I think that's a good point. If you're using AI to save two weeks of typing and you know what you're building, you can be way more effective with it than if you have no clue and building something without understanding what's going on

@raymcdermott Yeah, it’s 💯 the case that someone who didn’t know all the maths and hadn’t implemented this sort of thing before would have been unable to spec the work, evaluate it for correctness, &c. In this sense, Claude is sort of force multiplier for senior devs that kills jobs for the junior devs who we would want to grow into seniors. This is definitely bad. I’m also thinking about how having a tool that can, say, clone emacs in Rust in a week will lead to further fragmentation and bus factor one problems in the open source world because it reduces the natural incentives to collaborate. This also strikes me as a net negative for culture.

I think having claude code in my early years could also have helped me a lot... but perhaps coding wouldn't be as interesting to learn anymore... I'm not sure

I think having youtube and the plethora of videos would have been very helpful for my music skill in my youth. But then again, I would probably spend more time watching videos than actual practice. Maybe that translates to your thoughts, @borkdude, that we'd have less incentive for deep learning.

Update, as I review the big pile of Rust: it totally faked autodiff and the associated test 😂

hehehe