Ways to articulate policies on AI generated README files

Stream: Editors & UIs

[Castedo Ellerman]

How are people articulating their policies on what they read, or not, from others using GenAI to write README files?

To keep this topic focused, I specifically reference README files in Git repositories. Wikipedia says the README file "is intended to draw a user's attention to important and orientational information about the directory content".

The README file is not all documentation in general and it is not code. If people want to talk about AI code, perhaps start a new separate topic thread like this one:
https://leanprover.zulipchat.com/#narrow/channel/287929-mathlib4/topic/policy.20proposal.3A.20mention.20AI.20use.20in.20PR.20description.3F/with/539438023

Many of us, myself included, are struggling to decide what we read vs ignore that might be AI generated, such as Rob Pike with AI generated email.

[Castedo Ellerman]

A term that is new to me that I found on https://noslopgrenade.com is "slop grenade". It's not a neutral term, but it does make the point that using GenAI tech to produce lots of verbose material to read/look at does not make it better.

[Castedo Ellerman]

Jellyfin has a policy that states:

LLM output is expressly prohibited for any direct communication, including the following:

• issues or comments
• feature requests or comments
• pull request bodies or comments
• forum/chat/etc. posts or comments

and

An exception will be made for LLM-assisted translations if you are having trouble accurately conveying your intent in English

This doesn't address README files specifically, but I have always thought of README files as of the same spirit as the items listed above. They are intended for human reading.

[Castedo Ellerman]

The Jellyfin policy also goes into a bunch of detail in the section LLM-generated Tools, Clients, etc. Shared in the Community.

[Castedo Ellerman]

Some people might have a policy of "I don't read documentation (or README files) written with GenAI" But I would be a hypocrit if I had that personal policy because use GPT-4 with https://copyaid.it to proofread/spellcheck/copyedit documentation that I write, including README files. But it is to change existitng words, re-organize sententences, sometime big stuff like remove accidentally duplicated paragraphs.

To me these two uses of GenAI to write material for humans to read are very different. I'm not sure how best to articulate the difference. I currently call what I do "copyediting" for lack of a better term (and "copyAIditing" when I'm attemping to be funny). I don't know what to call the "slop grenade" style of GenAI writing.

[Patrick Massot]

What is the link between this channel and the topic of this thread?

[Castedo Ellerman]

Patrick Massot said:

What is the link between this channel and the topic of this thread?

The Lean-TUI README that maybe @Willem vanhulle wrote, maybe Claude wrote, I can't tell. It's just got lots of wierd content that is not making sense to me. It's kind of frustrating.

[Patrick Massot]

I see. I suggest we focus on specific relevant cases and do not start very general conversations that are not specific to Lean. This is simply not the place for those conversation. Note that I hate genAI and what it does to your world, but this is not the topic of this Zulip instance.

[Castedo Ellerman]

Patrick Massot said:

and do not start very general conversations that are not specific to Lean. This is simply not the place for those conversation.

That is reasonable. I see the logic in not having this topic here as it is not specific enought to Lean. Sorry for the distraction.

[Willem vanhulle]

Castedo Ellerman said:

Patrick Massot said:

What is the link between this channel and the topic of this thread?

The Lean-TUI README that maybe Willem vanhulle wrote, maybe Claude wrote, I can't tell. It's just got lots of wierd content that is not making sense to me. It's kind of frustrating.

Hey normally I wrote most of it. Parts of the diagram were generated in the past. Let me know if something is unclear.

[Castedo Ellerman]

I'm going to respect Patrick's request to not have a conversation in this Zulip channel about AI generated README files.

[Julian Berman]

I was also confused about this thread and what it had to do with Lean or editors initially, but I think the point is if you'd have asked "hey lean-tui's README is unclear because of X Y and Z" that is definitely on topic, what was confusing is what the general question of GenAI and READMEs has to do with Lean or this channel. Patrick can speak for himself I'm sure if he still disagrees, but I suspect no one will have a problem with that specific conversation.

[Castedo Ellerman]

AI generated anything feels controversial and I feel uncomfortable calling out a specific person/project. Which clearly I have failed to do here :face_palm: Sorry folks.

[Patrick Massot]

I’m sorry if this wasn’t clear. Julian is right, this is what I meant by “I suggest we focus on specific relevant cases”. Anything that is specifically about lean-tui is definitely on topic for this zulip.

[Willem vanhulle]

Castedo Ellerman zei:

AI generated anything feels controversial and I feel uncomfortable calling out a specific person/project. Which clearly I have failed to do here :face_palm: Sorry folks.

It is completely justified, but it is a bit strange that you mention the README as being AI-generated. For the past week almost all of the implementation was AI-generated, not the README. That is how I make quick prototypes these days :). However, I am going through every line now to check whether it actually make sense. I promise all of it will 100 % bio-degradable human by the end of the week.

[Castedo Ellerman]

Regarding the README file of lean-tui I will respond via Codeberg issues of the lean-tui repository. In hindsight I should brought this issue up on Codeberb rather than here in Zulip.