Rationale for filtering LLM usage links as off-topic
14 points by facundoolano
14 points by facundoolano
Hi, I've been following recent moderation changes and the discussion around them. I understand that the vibecoding tag was conflating multiple meanings and uses, and how moving to the building-LLM-tooling makes it more precise for that subset of the discussion. I also see how people may be tired of the LLM discussion and want to completely filter out anything LLM related.
What I don't quite understand and I'm not sure if it was clearly explained is why stories about "(not) using LLMs" are now considered completely off-topic in this site. There have been plenty of useful discussions around practices and figuring out how to adjust for professional programming in the presence of LLMs over the last few months, and the "flavor" of those discussions is very different from that found in other communities. Also, they seem to still be reasonably encompassed by the current "topicality" guidelines ("Will this improve the reader's next program", etc).
Considering they could still be filtered by anyone not wanting to see anything LLM-related, what is the rationale for not having something like llm-usage as a tag and keep those discussions on the site?
A lot of stories about LLM use boil down to either:
Stories of type (1) offer nothing useful to the reader, because you could literally replace "LLM" with "inexplicable magic black box" and not changing anything else. OK, so if I also have an inexplicable magic black box I could ask mine to do that. What did I learn about programming or computing from finding that out?
Stories of type (2) similarly offer nothing useful to the reader, because within days, perhaps hours, perhaps even by the time you finish reading the article, the LLM community will have moved on to a completely different set of Markdown files.
So what, precisely, is it that we're supposed to get out of an endless stream of stories about this?
There's no shortage of places where the topics you want to talk about are discussed. There's a clear shortage of places that don't drown out discussion with them.
Maybe I’m not part of the right communities on Discord or Mastadon or whatever, but in my life Lobsters has been the one place I’ve felt like my queasiness about AI tools (for lack of a better word) has been widely shared. I submitted an essay I wrote about that queasiness here earlier this year and it seemed to resonate with people. I was thrilled to know other people felt similarly, especially when I’m so in the minority at work.
Unless I’m misunderstanding something, if I wrote that essay today and submitted it then it would be considered off-topic under the de facto new moderation rules, since it was many thousands of words precisely about “(not) using llms.” That makes me a little sad, truth be told.
I’ve actually enjoyed Lobsters more recently without all of the non-technical AI culture war think-pieces like the one I wrote. I agree with others who have described it as “peaceful.” Keeping Lobsters focused on technical topics is also a valid moderation goal. But I also don’t know where else in my life I’ll now encounter the critical takes on AI from peers in my industry that previously I read here. I’m surprised at the lack of clarity from the mods on what to me feels like a seismic change, even if it might be the right one.
I was sceptical the new rules could change much but was thinking we should wait and see, and maybe improve them as time goes on.
A day or two afterwards, I realized I was finding lobste.rs pleasant and calming. It was due to the new rules. I was really surprised they had such an effect. I assume this is because users behave pretty well.
Are these rules perfect? Probably not. Do I enjoy the website more with them? Definitely.
I wasn't avoiding all LLM-related discussions or tagging or flagging; I took part in more than one discussion about them. Actually, I'd like some discussions to be possible again but in a much more delimited way. Maybe a recurrent "social changes in tech" just like there is a recurrent "who's hiring".
It’s a bit hard to tell but I think you’re referring to non-technical thinkpieces about the impact of LLMs on the workplace or FOSS communities and ways of functioning in those spaces while using LLMs either more or less than the norm? If so I find that style of post basically worthless. The half-life of their relevance can be counted in months if not weeks.
A recurring theme of the LLM discourse on this site has revolved around people wanting something like “a sane place to discuss LLM $TOPIC” which this site is perceived to provide better than other sites. Which clearly ignores second-order effects. It’s like if I wanted to discuss the finer points of cryptocurrency consensus protocol design in mid-2021.
I think the following statements are logically consistent:
From the About page -
Some rules of thumb for great stories to submit: Will this improve the reader's next program? Will it deepen their understanding of their last program? Will it be more interesting in five or ten years? Is this compelling material that will seed a discussion, or is it just an artifact?
Mod @355E3B comments as mod -
The litmus test I'm using for this tag [
ai-tooling, since renamed tobuilding-ai-tooling] is if the body of work still makes sense if I were to replace any instance of LLM with consultant and it still hold up, it's more likely than not outside of the bounds of topicality for the site.
To me, it seems like the mod team has come down pretty hard on the side of programming as a human activity, and against using GenAI (and possibly future technologies) to produce software artifacts.
There have been plenty of useful discussions around practices and figuring out how to adjust for professional programming in the presence of LLMs over the last few months […] they seem to still be reasonably encompassed by the current "topicality" guidelines
I'm not so sure? From the About page's section on Topicality (my emphasis):
Some things that are off-topic here but popular on larger, similar sites: entrepreneurship, management, careers
Just because it's good or useful doesn't mean Lobsters is the right place for it. I don't have the rationale for why the guidelines say what they do, but the guidelines are not as simple as "would improve a program" implies "is on topic".
I shared a story that fell foul of this the other day. The October inflection point for vibe game development talks about how Opus 5.5 (launched September 22nd but its capabilities started being fully understood over the month of October) can now do 3D modeling, music composition, and game design - you can tell it "make this more fun" and it knows how to improve a game to make it more fun. I supported that claim with four examples of my own game experiments (prompts and transcripts included.)
It was flagged off-topic and, after re-reading this comment, I concluded that the flagging was indeed in the spirit of the new rules.