hi i'm mikus
you probably know me for a project or two, which is alright. this website (and anything under this domain) is more or less related to misc projects outside of my "professional" life and mostly focuses on fun experimental tasks, such as integrating ai into life in a non-bullshit way.
llm api @ llm.cappu.ch
osu!lazer web port @ osu.cappu.ch (pretty glitchy and buggy and doesnt load on some devices for some reason)
a small rant
personally, even though i like to do mle/mlr, i don't like slop (see try.cloudflare.com back on the 16th of september 2026). now, my definition of slop varies quite a bit. of course, something like language models generating code for video game modding wouldn't count as slop under my personal definition, but it probably does under someone else's. something thats created by a generative model isn't inherently slop, but it can be if the content that was generated isn't processed afterwards or used in some other capacity, with human input. so generally, slop is anything that doesn't have a human touch or a moderate amount of effort put into it, not 100% generated by a mathematical model.
i'm personally anti-vibe coding, and i've quite literally gone through a cognitive decline because of sonnet 3.5 => opus 4.5 (truly rotted my brain), so i've taken a more disgusted approach towards the topic. (without harassing anyone, duh) if you actually test / qa the code yourself, i have no issues with that, as long as your readme doesn't have rocketships plastered all over it.
it sounds a bit ironic to see a literal llm api on this website, but intelligence should be democratized and made as cheap as technologically possible, as realistically speaking, ai (not specifically llms) will probably change the world. (for better or for worse, ideally for the better though)
on top of all of this, i specifically don't like the use video, image and audio generation models. it's hard to explain, but the best way to put it is that it fits with my definition of slop. i like some ai cat slop gifs (i literally have one up in the header), but i still find the issues with the existence of these models. it's somewhat of a "necessary evil" approach to ai, where the models learn to generate art and content that these models learn and eventually generalize.
that's not to say that llms don't have any flaws. i truly, at the bottom of my heart, hate anything relating to language models generating literature. that's another ability that eventually comes from generalization. llms for fact checking (using tools!), llms for grammar correction and spelling correction, are great use cases for llms in the process of writing literature, and that's about it.
tldr, i find that ai is used best as a tool, and not as a replacement. i don't want to outsource my thinking, creativity and taste to a massive set of matrices.
and obviously, this goes without saying that using ai shouldn't invalidate whatever you're doing. using ai to prototype, understand content, check something, automating tedious work is a massive difference from shitting out an entire full fledged product for you. the thing i care about is that a human is there, making the deliberate actions somewhere in that process.
i also don't really like the idea that being critical of ai means you have to reject the technology in its entirety. that's just absolutism, and i think it's a terribly common mistake. you can think that llms are useful while also thinking that flooding the internet with generated articles, images, music, videos and code is miserable. (which is 100% true and you know it) these aren't conflicting positions unless you maintain that "ai" is one giant homogenous thing.
ultimately i want computers to make people more capable. automate the boring and annoying shit. help people learn things. make information easier to access. make tools cheaper. let people experiment with things they otherwise couldn't, like writing code.
and since i'm already ranting, the environmental discussion around ai is also just as terrible. i'm not downplaying it, or saying it doesnt happen, and training and inference does use a fuckton of energy and water, which it obviously fucking does but the nuumbers get stripped of every bit of context imagineable, especially when talking about water usage.
you'll see some absurdly precise claim about how asking chatgpt a question uses "x bottles of water" as if there's a bottle of evian being fucking poured directly onto a gpu. datacenter water usage depends on where the datacenter is, how it's cooled and the weather, grid, hardware, util and wether you're talking about water withdrawl or actual consumption. turning all of these variables into one universal number per prompt is incredibly misleading.
if anyone ever managed to implement generative ai into geometry dash in a meaningful way i would probably fully unironically start calling for their death/suicide
— callie , 🏳️⚧️ (@QalliCallie) August 20, 2025
and MY god, if your opinion on a technology has reached the point where you're sincerely calling for the deaths of people who implement it into a FUCKING rhythm game, maybe the discourse has completely lost the plot? there's critism, there's hatred then there's this shit. you can think that generative ai is creatively bankrupt, environmentally wasteful, exploitative of artists, annoying or all of the above without deciding that someone experimenting with it deserves to fucking die. at that point you're looking for a morally acceptable target to direct a shitstorm towards.
and the opposite side of the discourse is hardly any better, there are people who treat any critism of ai as proof that "oh but you don't understand the technology", as if understanding what a transformer is and what it does settles every ethical, artistic and social question surrounding it. understanding how something works doesn't require you to praise every possible application of it.
this is probably the part of the ai discourse that pisses me off the most. everything has to be absolute, either ai is the greatest invention in human history and anyone who challenges that idea is a technologically illiterate moron, or it's an irredeemable plague upon humanity and anyone who touches it is akin to nestle's ceo and destroying art.
maybe, just maybe, a technology can be extremely useful, interesting, occasionally fucking awful, and worthy of criticism at the same exact time. can we take pride in how far humanity has come to develop near-super intelligent machines out of literal stones? we invented lasers, shooting light at near atom-level precision and we choose to fucking bicker about it?
all in all, it literally comes down to just one thing, don't automate the part that made the thing worth making in the first place.