I've noticed that lately a fair number of topics have been started regarding AI usage in programming (like anyoftheseposts), with many more unrelated posts that turn into discussions about AI.
In today's world, AI obviously has to be addressed, and I think the official stance on LLM usage in the site is clear and good. I was just wondering if we should be looking into policies regarding the discussion of AI.
On one hand, people want to discuss how to use AI when programming Rust, and I don't think that's necessarily against the spirit of this forum. But to me, it's gotten to the point where I no longer consider the discussions constructive in any way. Nothing new is being said in any of them, and they almost always end up far off-topic, with very few references back to Rust. (I acknowledge my participation in many of these discussions; that's on me.)
I'm not advocating any particular course of action, simply asking what the mods and community members think of the subject. I just find the constant LLM-focused posts almost as tiring as the constant LLM-generated posts, and roughly as constructive. Even worse because they're so easy to get drawn into for both sides of the debate.
I’d like that, insofar as there are discussions of AI (I take no position on the desirable quantity of such topics), that they stay on topic. For example, the recent thread SKILLS.md for Rust development was asking a concrete question about using AI tools — well, mostly concrete, being a “How do you, personally, do this?” question. “I have found this is not feasible” is a fair answer to the question asked, but arguing about what an LLM is capable of in principle, whether positively or negatively, should be promptly cut off as off-topic for that thread (and, as you have noted, often tediously repetitive).
As the author of that thread, I completely understand your perspective, and I'd like to say sorry if it caused any clutter or fatigue in the forum. Being relatively new here, I'm still learning the implicit etiquette and norms of this wonderful community.
To be honest, I didn't expect the post to spark such a massive wave of discussion. I guess many questions surrounding AI naturally trigger divergent thinking—we often start with an engineering problem but somehow end up having a philosophical debate.
I think a great solution moving forward could be creating a dedicated sub-category or a specific tag (like #ai-discussions) just for these topics. That way, those who are interested can dive deep into the "philosophy," while keeping the main technical sections pure and heavily focused on Rust itself.
Thanks for the guidance, and I'll definitely keep this in mind!
I don't blame you for your post, like kpreid said that was a pretty concrete question. That it devolved into unhelpful argument (again, my involvement is my bad) is not your fault. That said, it still serves as an example of how even well-intentioned, rule-abiding, etiquette-following posts about AI tend to go poorly.
That has some merit. For those of us who wouldn't want to see AI-related posts, there's an option in the preferences (under "Tracking") to ignore certain categories. However, that wouldn't help with topics that get off-track and devolve into AI discussions (I don't have links to any, but I believe there's been a couple lately). Perhaps that could be managed by splitting out AI derailings into new, properly-categorized topics?
It's a dangerous slope. If you transpose that to a question about the language, and if someone's asking "how you personally do something", the something being unsafe or not idiomatic at all, a response along the lines of the compiler not being able to do that well, or Rust not being designed that way, shouldn't be cut off as OOT. Yet it's following the same pattern.
Explaining why something isn't feasible is a valid answer, possibly more useful than a "I've found this isn't feasible". At least in the mind of the one who answers.
I think the negative reaction to using LLMs is mostly twofold:
Some developers feel threatened by the technology or are under pressure to use it, and the topic puts them ill at ease.
Some developers thing that those tools are inappropriate and explain it to the OP.
Those reactions are based on grounds that cannot simply be dismissed, and the arguments put forward aren't necessarily known to the OP (after all, they're asking a question).
I can't think of a solution off the top of my head, but I don't believe the issue can be solved by silencing people who genuinely think something's wrong.
Of course, it all depends on what exactly the question is. In the threads linked above, two are vague: asking about vibe coding or whether we should still write code manually can't possibly avoid the debate; I can't think of an algorithmic answer or a methodology to do vibe coding. The third one explicitly asks for advice.
Maybe there should be a restriction on what to ask / how to ask it, and a subsequent restriction on what to answer to that, but I'm not sure it's easy to enforce. Another idea is to have a dedicated part of the forum for those specific questions, (but, to be honest, I'd very much like a feature to optionally hide those discussions entirely, and so might others).
For what it's worth, it can be somewhat of a controversial topic, depending on the exact stance and/or unvoiced assumptions baked into the given thread's topic. Is the OP attempting to leverage an LLM to write better code themselves? Might be a conversation to have. To build their own virtual SWE that would churn out 100 crates a day, neither of which is properly reviewed beforehand, polluting the ecosystem, then promptly moving on towards the newer/shinier lands without a second thought? I don't think the community has any duty to absorb the impact of anything of the sort.
"AI" itself has become a catch-all term for at least a dozen of different technologies: GPT's w/o RAG, "vibe coding", machine/deep learning, "prompt engineering", diffusion models, agentic coding, snippet generation/auto-complete, blind delegation of any/all responsibility to a third-party cloud-hosted "PhD intelligence", which is advanced enough to generate 30K+ LoC per day despite its routine inability to tell the number of R's in the word like "strawberry". What do we do with that?
I wouldn't be outright against prohibiting the use of "AI" itself as a keyword, forcing the topic to steer towards the specifics of what the OP actually means when he talks about "AI", long before the button "Post" becomes active. Are they using an LLM? Built from scratch? Someone else's? Have they ever built their own neural network? Do they know just how little a GPT's "hallucination" has to do with its equivalent in human psychology? Or do they think they cajole their ChatGPT assistant into an "expert mode" with yet another "do not make any mistakes" line at the end?
If this forum is to remain centered around Rust the programming language, its focus must pertain to Rust the programming language. Not the ways around the language for someone who would rather never write a single line of code in their life come forward, if they can help it. Not the best prompts or skills or markdown specs for the language to be fed into any particular agent/model at the very forefront of the hype cycle of the day or the week or the month. Not all the variations of the exact same "I'd really like to get my machine to do X, using this great language you've got going here, without putting in a single ounce of effort into bringing myself up to the level of expertise and knowledge and understanding that seems to be required to do the thing on my own".
I don't think there's any way to test people before making their post, either. Forcing people to look up Ashby's Law however, before posting anything AI-related, wouldn't go terribly amiss.
I do read these discussions, and appreciate the wide range of views and experiences. I do not contribute often as my own experience is nil, even if I have some thoughts. I don't have any objection to these discussions at the current level.
AI, LLM's whatever are having a huge impact on the world of programming. As such I think it's nice that programmers have a place to discuss it. This is such a place by virtue of it being for programmers. Naturally there will be a lot of repetition of opinions and seemingly endless debate. I suspect that will subside over time.
I think I've gained some relevant new perspectives from the LLM discussions on these forums, so I'm overall thankful for them. With that said, they seem to pop up a little more frequently than is useful and they seem to end up in the same old loops eventually.
In general, I think I'm more tolerant with regards to repeating posts/topics than many others. Basically I feel that:
New users haven't read the old discussions, and even if they try to use the search functions to find what they are looking for, it's not always they'll find the old discussions.
As new information emerges over the years, sometimes old discussions are no longer relevant.
People who engage in topics they already have engaged in, tend to rephrase things in ways that can be extremely helpful sometimes.
But at times I feel that the LLM discussions happen too frequently to fall into any of these categories.
Not sure what to do about it through. As you, and others, have said -- it's an important topic.
I don't think that's a very helpful answer. It doesn't solve the more general problem of the dilution of the purpose of the forum, nor does it help at all with posts-gone-awry that used to be informational, but aren't anymore.
I had the same thought. Other types of side discussions happen already (e.g., discussions comparing programming languages), and sometimes mods split them into a separate topic. Seems like the same thing could happen with AI for the same reasons. In other words, I'd leave it up to the mods.
