i've read basically everything today, so instead of replying thread by thread (again) i'll just say it all here once for the last time..
people criticising us is fine, we published a whole document exactly so people could go through it. what upset me is seeing months of work reduced to one word by people who never opened that document, over an amount of generative ai that in this project is tiny. what a model did for me was explain things i hadn't seen before, help me when i was stuck on a bug, and write out repetitive blocks i already understood. and i said all of that myself in the document, way before anyone screenshotted it.
and let me be clear on one thing, because it's going around as if it were a quote from me. osc is NOT vibecoded, and i have never once said it is. the document has a section called "it's vibecoded" because that's the accusation the section answers, and the first line under that title says i'm one of the first people to be against vibe coding. reading a header as a confession and skipping everything written under it isn't reading the document. i didn't make up the definition either, it's the one from the person who coined the term, and it means letting a model write code you never read and can't explain. i can explain every part of this, because i had to build every part of it.
so, this is what changes.
from now on i'm not using generative ai to help me develop osc, at all. no llms, no code assistants, nothing.
and i want to be straight about why, because it isn't that i've decided i was wrong. it's a tool. every programmer and engineer i've talked to about this uses one, and it isn't only the people i know, stack overflow ran their developer survey last year and 84% of developers said they use ai tools or plan to, with about half of the professional ones using them every single day. what matters to me is what you give it to do. your art? no. the whole project, so that you ship code you can't read? no. asking it to explain a concept or help you track down a bug is using a tool. and to be clear, i'm not putting ai and stack overflow on the same level, they're not the same thing. what's the same is how it's being treated. copying an answer off stack overflow got called cheating and not real programming for years, and the same was said about ides and autocomplete before that. every time a step gets easier, the first reaction is that whoever uses it doesn't really know what he's doing.
and on the environmental side, which is the part of the criticism i take most seriously, the iea put data centres at around 1.5% of the world's electricity in 2024 and expects that to more than double by 2030, and that growth is a buildout, decided by a handful of companies choosing where to put those buildings, how to cool them and what powers them. google published its own measurement of a single text prompt and got about a quarter of a watt hour, roughly a tv left on for nine seconds. the damage is real and worth being angry about, it just isn't caused by one person asking why his server crashed, and boycotting that person doesn't touch a single data centre.
but i'm stopping anyway. because a lot of you aren't comfortable with it, and that matters more to me than it being faster for me, and faster is the only reason i ever used it, one person building a whole platform against a deadline, which isn't a good enough reason when the community feels this way about it.
and we're not stopping there. we're going to look at removing the last two ai pieces running inside the platform too, and i'd rather name them properly than leave it vague. one is the external moderation check, which isn't a chatbot, it's a classifier that reads a piece of text and gives back risk scores, and it only ever got the few borderline cases our own local checks couldn't call. the other one is deepseek, an actual llm, and i think people are picturing something much bigger than what it does. it does the searching for us. when we need to check whether a show's title or description is actually correct, or look into a creator someone warned us about, it goes through wikis and public pages and hands us back the sources. then we read them and we decide. none of it was ever automatic, and none of it ever would have been.
taking those out isn't free though, and i'd rather tell you now than after. it means doing all that searching and all that moderating by hand again, with a handful of volunteers, so things will be slower, and i can't promise everything stays exactly as good as it is today. we'll find another way to keep up. and if something does get worse, you'll hear it from us, not from a screenshot.
and all of this gets written down, not just said here. the transparency document gets updated, the site policies get updated, and on kickstarter i'll change everything that's still editable now that the campaign is live. i'd rather the documents say exactly what we do, so that nobody has to trust a tweet.
and if you don't want to take my word for any of this, you don't have to, because my word isn't the only thing holding it up. the eu ai act's transparency rules start applying on the 2nd of august 2026, which is today, and under them any ai a user actually touches has to be declared and labelled, and we're in italy, so that's the law we answer to. kickstarter's own rules already made us fill in an ai disclosure before this campaign was allowed to go live. our privacy policy has to name every outside service that ever touches your data, so quietly keeping one wouldn't be a bad look, it would be a gdpr breach the italian data protection authority can act on. and the independent security audit this campaign pays for reads the actual code. so if i said this today and did the opposite tomorrow, it wouldn't just make me a liar in front of you, it would be a false claim made to people who paid money because of it, and that one has a name in the law.
osc was built for you and not for me, so if this is what you want, it's the least we can do.
short version, so nobody has to guess what i meant. osc is not vibecoded, and i never said it was. from today, no generative ai in developing osc, at all. we're also going to look at removing the external moderation check and deepseek from the platform. the transparency document, the site policies and the kickstarter page get updated to match. i don't think i did anything wrong, i'm doing it because you asked. and none of this is just a promise, the ai act, kickstarter's rules, our privacy policy and the external audit all make it checkable.
even shorter version, no more generative ai :)
-Dan <3
Sometimes I forget that Angelis actually narrated railway series stories, and that’s how things like this come along!
Down The Mine and Paint Pots And Queens double feature!
...And here is the semi full clip since Twitter won't let me upload more than 2:20 minute long videos :/
Special credits to:
𝙈𝙖𝙜𝙞𝙘𝙍𝙖𝙞𝙡𝙧𝙤𝙖𝙙𝙉𝙚𝙧𝙙
𝙋𝙪𝙜𝟮𝟱𝟵𝟭𝟬 on the Archive
𝙂.𝙒. 𝙎𝙝𝙪𝙣𝙩𝙚𝙧 and other 2 people which I lost track to for the SFX.
@CPImagined_ I'm working on a special project for the shutdown also if you can, can you give me a date when the game will shutdown so i have a deadline. Also left a image of the rough draft of the project.