Proposing a novel legal attack on LLM-generated source code where you claim it is incompatible with the GPL because the prompts are the source code and the emitted C or Rust or whatever is a compiled binary
@mcc i will not turn this into a perl joke, i will not turn this into a perl joke
@EndlessMason @dvshkn I'm a perl coder, everything i write is write-only
@mcc I mean that's consistent with the boosters' position that you don't need to understand the output
@zwol musk and others are literally arguing that source code is now "assembly language" and that understanding it is exactly as useful as knowing asm and for the same reasons.
@mcc I hadn't seen that specific argument, but that's very funny to me as a person who used to teach college sophomores to read assembly language
@zwol I don't think the people making the argument are arguing nobody learns, or that nobody gets use in a professional context out of knowing, assembly language. They understand that comes up. In fact that is probably what makes the argument pernicious, that they can explain away counter-evidence by going "aha, that's one of the exceptions I've already outlined"
Imagine if you designed a programming language where the compiler needed to make it work only existed temporarily at the moment the program was compiled and permanently deleted itself when compilation were done. Would such code be compatible with the GPL? Because that is effectively the situation with a system with stochastic semantics, as a code assistant/agent prompt system uses.
Note: The above posts are "joshing" and not meant to be interpreted as something I think is a good idea
@mcc this isn't completely ridiculous. Many people do take the high level language analogy seriously and check in the prompts. I wonder how GPL applies to obfuscated code (are you allowed to deliver minified, obfuscated js without the original when the original is based on a gpl'd product? I'm guessing not). Sufficiently sloppified code is indistinguishable from obfuscation.
@thomasjwebb I didn't say I thought it was a BAD idea, either
@mcc Yet it might actually work.
@mcc @cwebber after scrolling by, came back: this is actually a novel way of thinking about ai interactions. Prompts are written expressions.