@filippo @whitequark @untitaker

If these were generated by an LLM the prompter probably didn't even think about how much spam they were creating. The AI probably did all the work of posting these. The user may not even know they got posted at all.

It is becoming very unbalanced because the tools were developed for slow humans, but are being used by fast AI bots.

@poleguy @filippo @whitequark @untitaker They were. The poster states it very clearly:

"It's totally "vibe-coded", I haven't even bothered to look at it - but I would guess it's in a better state than the original project."

that is, uh, not exactly pro-social behavior in my book.

@poleguy @filippo @whitequark @dave_andersen tbf the context about this being vibecoded and vibeposted is not at all apparent from Filippos post

@untitaker @poleguy @filippo @dave_andersen yeah, this completely changes how I feel about its appropriateness..

@whitequark @untitaker @filippo @dave_andersen

I had noticed the text: "(This post has been created with the h..." in the original post, so it was clear to me that it was an AI bot. Also, no real human would do this.

But yes, the tide has turned. We can no longer presume good intention. AI bots do not have "intention". They just do as they are programmed.

@whitequark @untitaker @poleguy @dave_andersen I generally agree with Russ Cox in https://groups.google.com/g/golang-dev/c/4Li4Ovd_ehE/m/8L9s_jq4BAAJ that the tools we use don't change the responsibility we need to take for the results, in one direction or in another.

@filippo @whitequark @untitaker @dave_andersen I read that whole response. I agree.

My first joking comment (to prove I read it) was going to be that there is a single character typo: "Now would be a good time too." That 'too' should be 'to'. But upon re-reading, it seems he really did mean 'too.'

But I think the focus needs to be upon velocity, not copyright. Low effort submissions used to be quick to respond to: they wouldn't have great code. Now they seem to have great code.

@filippo @untitaker @poleguy @dave_andersen to express myself clearer: I don't think "going through the issues of a project not maintained for 4 years to say they've been fixed in a fork elsewhere" in and of itself represents spam; it is relevant communication of clear potential value to its recipients, who opted into hearing it at one point or another. (one could debate whether a PR should have been opened first, but that's getting into "splitting hairs" territory.)

I do however consider LLM contributions themselves spam unless demonstrated otherwise against a high, clear bar

@untitaker @whitequark @dave_andersen @filippo

That's spot on. An author not taking responsibility for their submission (I don't have time, etc.), especially if it's non trivial in length (aka: AI)?

Then we need a high-gain feedback mechanism to discourages this.

Maybe those submissions should go to the bottom of a deep queue.

If the human author really cares, they will engage. Otherwise the AI submits will begin to turn to petroleum over the eons.

Use a big Integral term in the PI loop.

@whitequark @filippo @untitaker @dave_andersen

However the right thing to do here would be for the author to share the prompt that they gave to the AI. It was probably only three sentences.

Possibly this rule would solve a lot of our AI trouble: all AI submissions must include all prompts. (That's the 'source code' after-all.)