#Rules on AI

62 messages · Page 1 of 1 (latest)

jolly notch
#

I suggest the following new rules:

  1. If your message, image, or code is generated by AI, you have to say that.

  2. You're not allowed to send messages or code posted here to AI unless the person who wrote it agrees.

I think it is fairly obvious why this is a good idea

—
Update

Commenters have suggested the following rule:
Don't respond to questions with AI answers

Commenters have noted that rule 2 is difficult to enforce

Commenters have noted that this may already be implied by the etiquette guidelines (not the rules)

#

To specify further,

  1. By "AI" I mean generative AI that imitates humans, like LLMs or diffusion models, and that processes sensitive information, like face recognition or medical diagnosis.

  2. By "generated by AI", I don't mean any code or content that contains anything that was output by an AI. The (A) core element(s) of the message has to be in large part recognizably similar to the original output from the AI.

latent nebula
#

2 seems hard to enforce. I'd say it would be more productive to give a blanket "Don't copy/paste AI answers when helping people", since that's the main time AI content shouldn't be sent.

jolly notch
#

I generally agree with "Don’t copy/paste AI answers when helping people", but I thought that’d be a harder sell.

#

Ideally I’d want your rule as a third rule…

#

I think "Mark AI generated content as such" and "Don’t respond to questions with purely AI-generated answers immediately" could be ok

latent nebula
#

Coming from the Python Discord where that is a rule, I don't see why it would be a hard sell, especially since AIs work worse with rust than with python.

jolly notch
#

I wasn’t sure how the consensus was on that

#

I am very convinced it is almost never a good idea to take anything an AI says seriously without a lot of context and that it is almost never a good idea to ask AI for help with a specific problem or immediately

#

but I don’t know how many people agree

granite magnet
#

2 is "enforced" in #etiquette-guidelines no. 2

jolly notch
#

I think it could be expanded on

#

and maybe made a rule

granite magnet
#

just call a mod whether you suspect AI is used like that

granite magnet
#

i think it was a discussion here

#

in server feedbacks

jolly notch
#

ok

jolly notch
#

I searched for "AI" here before opening this thread

dawn temple
hearty igloo
#

Love this idea

#

we've also had some antisocial ai stuff come up and its rubbed people the wrong way in general

jolly notch
#

I would then suggest this specific wording:

Rules for AI

If your message, image, or code is generated by AI, you have to say that.

This applies to AI like ChatGPT or Stable Diffusion, which make new content in a human-like way.

But if your message is only based on AI output the AI part isn't mostly unchanged, it is allowed.

—

You're not allowed to send messages or code posted here to AI unless the person who wrote it agrees.

This also applies to AI dealing with sensitive information, like face recognition and medical diagnostics.

Some people are concerned with the privacy of their data. Some people don't want to train or help AI companies for free.

—

Don't answer questions or help with problems by just copying AI output.

AI generated answers often contain errors, and people often prefer talking to humans about their problems.

It's also usually unnecessary, because AI tools are widely available.

long meadow
#

Dude, have you heard about AI-first IDEs? Think of Zed or Cursor. AI is literally everywhere now—translators, phones, email, you name it. Honestly, it’d be pretty weird to feel like you have to announce you used AI. That’s as ridiculous as saying, 'Hey, just a heads-up, I wiped my butt with this hand,' before shaking hands with someone! ferristhumbsup

feral badge
# jolly notch I would then suggest this specific wording: ### Rules for AI If your message, ...

Generally, I agree with these rules, and would like to adopt them officially. After polishing up the wording a bit, that is, but we can do that in #mod-chat.
(To be honest, this kindof reflects existing informal policy.)

To be clear, the section on privacy and not sharing peoples' data with AI without their consent is already a rule we enforce here, although we don't have too much power over that.

We also rule against misinformation separately already, but yeah.

#

The wording I want to fix up is this:

But if your message is only based on AI output the AI part isn't mostly unchanged, it is allowed.
I get what this is saying, but it's, in my opinion, more difficult to parse than it needs to be.

dawn temple
#

can we s/AI/Generative LLM AIs for clarity

feral badge
#

There are already examples at the start of the proposed rule, which I think I'd like to keep.

dawn temple
#

thats fine too

jolly notch
# feral badge The wording I want to fix up is this: > But if your message is only based on AI ...

consider instead

the above shall not apply in cases where content derived from AI output has no core element substantially derived from AI output, where a core element is any part or aspect of the work that is particularly relevant to ordinary observers or observers skilled in the art, or is an aspect judged by the ordinary observers to be prominent with respect to other aspects, and a work is not substantially derived when the creative decisions in creating the work succeeding the generation by AI in large part overwrite or in skill or number substantially exceed those necessary for the creation of a similar work as was generated by AI without the use of AI, or transform the AI output as to fundamentally alter its purpose or fundamental character

#

this is a joke btw

#

I swear to God that wasn't AI-generated

feral badge
#

nice lol

hearty igloo
#

ah crap it probably was AI'd then hand editted, darn you schuelermine

image.png
jolly notch
#

fuck I made a mistake

#

fixed

dawn temple
jolly notch
#

I don’t think this needs to be specified

feral badge
#

Sure, default to placing a disclaimer if you're not sure.

Now, if we're talking "do you actually need to", that'll depend on whether you can process what you've made and be like "yes this is something I would write" and "there are no 'facts' stated in here for which the only source I have is the LLM".

#

Generally though if you push it, we're just going to rule that you need to place the disclaimer more often.

#

And, we all do know what a "disclaimer" is, right?

#

A "disclaimer" here is just a statement where you disclose "yeah I didn't make <this thing or this part of this message>, it came from <this generative AI thing>"

#

you can word that however you want as long as the idea comes through

#

looking forward to someone noticing "hey hang on Monad that's normal citation practice"

stable karma
#

speaking entirely for myself here but personally the main things i'd like to avoid are (1) feeling like you're talking to an ai and (2) answers to people's questions being based on llm output.

if you're asking a question and using an llm to figure out the right terminology for something then that's whatever as far as i'm concerned, as long as the message itself is your own words. but answers i feel have to be held to a higher standard

feral badge
#

I personally care more about the second thing (2) there, but yeah your aims here are agreeable

#

the first one is sometimes an inevitability, just because the occasional non-english-speaker will pop up and be using, like, Google Translate, and that sometimes comes off a bit... stilted?

#

but yeah.

stable karma
#

yeah i was about to say, i think some leniency can be afforded to people who struggle with the language and need some kind of translator tool to make themselves understood

#

though it's also worth considering that it can sometimes be difficult/frustrating to get your point across when talking via an automatic translator, so in those cases it might be more productive to try to direct them to somewhere they can use their native language if possible

#

but those places aren't always available and i also don't want to show people the door just because they struggle with english, so idk

dawn temple
#

why not

brittle cosmos
feral badge
random apex
#

Personally I would prefer such a poem to disclaim its source

brittle cosmos
#

If it were made by a human author you would have to disclaim the information as well. Now I do wonder though, do I have to disclaim an AI generated meme template?