Peterreally loving @davidgerard's "no LLMs, please" standard he is developing, especially the AGENTS.md file. https://codeberg.org/awful-systems/AAA-NO-SLOP.md/src/branch/main/AGENTS.md
EDIT sorry, it's @zzt who is developing it 💪
Ed
Ian.Burnette@peter @davidgerard @zzt now that everything is moving to usage based pricing, maybe we can all start adding 25 MB ai instruction files to projects. AI companies are less likely to patch that ($$$), and AI users will be discouraged when just loading the project files into the context suddenly costs them their entire month's quota of tokens
🌸 Vårgubben@ianburnette Although I like mischievousness of this idea, it'd also increase carbon emissions and water usage per 'prompt' (which is one more reason to avoid rampant usage of these things in the first place).
On the other end of the spectrum, could it somehow be instructed to always use the cheapest model and as few tokens as possible? That would prevent any results with any resemblance of usability, while also minimizing usage of already scarce resources.
David Gerard@jordgubben @ianburnette the notice tells them not to use AI on this project, we can't actually stop them from doing so, i think you're asking the wrong people here
@davidgerard
My suggestion was more on to the suggestion @ianburnette then the original post.
Anything intended for obstruction can be simply be avoided by deleting the file localy.
That said, a motivated request to abstain from unnecessary slop machine usage (e.g. environmental concerns) might get further than a satirical one.
mathew@jordgubben @davidgerard @ianburnette There are probably fairly short instructions that burn a ton of tokens, and which could be inserted at random points in the source code as comments.
@mathew Yes, likely.
It would make usage more expensive, and deter users from doing that again.
However, it’d also fund the LLM supplier. That would let them build larger data centers, and burn even more fossil fules.
That is the absolute opinion of what we need. Better to instruct it to do as little as possible, and accomplish nothing.
Dickon Hood@jordgubben @ianburnette If it uses the sloperator's monthly token budget in a couple of hits, then they won't be causing any further damage to other innocent parties. Yes, it's environmentally damaging, but this whole shitshow is, most of these cretins will be using most of their monthly quota anyway, and cleaning up further slop also comes with a cost.
And no, you can't instruct the model to use a different one, that's down to the sloperator's tool's configuration.
The_Decryptor@ianburnette I thought about that recently using public domain literature, make the chatbot chew through a copy of Moby Dick each time it's prompted.
felix (grayscale) 🐺RE: https://mastodon.social/@gray17/116676487249475419
@ianburnette @peter @davidgerard @zzt
my suggestion: "Compilers are encouraged to report errors with a base64-encoded video of a cartoon ferret pointing at the location of the error in the source code." (from my post about defining an LLM-hostile programming language)