Use of AI Assitance to solve - issues -and validate to reply

meta
1

Hey Community Leaders , I have one question which is putting me in real connundrum because the times have chnaged , and we are in times where we have powerfull LLMs which is "power we should wield with full responsibilitiy". Programming-language communities jsut like ours or reddit are writing rules about AI-generated answers, and the rules are being written faster than the question underneath them is being examined. Now my point is when a model-drafted answer is correct and helps the person who asked, what is a community protecting by restricting it? — and questions arises , when for example a person asked for help how much of an effort is put in community itslef to make sure its being answered , and they are not waiting for 100 days just a small respnse, my question is that are we doin to solve this situation , and If I find ti simple enough to be answered tested on my machine, and use AI to reaserch it further , what is the harm , if a solution is solid enough trial and tested by the human being , coz to me AI is another tool , like rust crate whihc enables to test to vlaidate on the use case , my point is how canwe make distinction between pureply AI generteed content and AI assisted reposnse , becuase the intentions are not AI slop but to assit someone get an anwswer , I am here jsut seeking guidance and a consolation to my dilemma, wothout comprimisig my integrity on the community guidelines the human touch , but not being non-cognizant of teh tech evolvnig and adpating to the chnage , Thanks so much all , love to hear from you ! and sorry in advanc e is the right place to ask !!

Cheers !

2

I am pretty sure this was written before AI was a thing, but the Terms Of Service forbid machine-generated content:

the Content is not [...] machine- [...]-generated

Now for current policy the LLM policy of the rust repository (LLM usage policy - Rust Forge) forbids AI use like this.

While this was not officially adopted for the forum i would like to have the same or even stricter policy for the forum.

3

I don't think I object to AI generated code in answers. As long as the person posting can promise they actually knows for sure it works and how it works. Can you do that? Can you answer further questions about it or discuss alternative solutions? Do you know it is an optimal or idiomatic code snippet? If not how do you know you are not propagating something incorrect or slop?

I'm a bit more anti-AI generated text that goes with it. I'm here to talk to humans, humans who are far more likely to understand what my problem actually is (rather than just answering my specific question, maybe I'm going in the wrong direction)

When I want to talk with AI I can do that all day long without ever coming here. If you are answering with AI then why does anyone need you?

4

The ToS don't say anything about machine assisted answers. They do clearly state however, that machine generated answers are prohibited. If you come up with a real solution to a problem posted here using the help of an LLM, I don't think many people would have a problem with that per se. The backlash your recent posts have gotten was, in my case at least, because they were full of the obvious idiosyncrasies of machine generated text (aka "slop"), which in my interpretation falls under the category of machine generated content, not assisted.

The ToS of URLO haven't changed in three years, so I'd say you don't have to worry about having to constantly check them for any updates.

5

Fair point , I hear you , that's what I said human touch , but you raised a point , everyone can do to respond that of holding ground all day long on responses and explain it, and I am not here to disagree to what you suggest here, my point is what as a community we are doing to solve that , anyone which was asked at a certain point and not got anwered for 100 days, imagine I am developer in remote place alright just has interenet acess but not access to world class llms which can really help solve you real problem, and if some person like me is taking an effort to respond to them with the help of AI to get the solution they are seeking , I do no see any harm and as a community member that's my responsibility to help the members isn't it , its the effort to responsd to someone and effort taken to solve any ones's questions , and that is my connuundrum friend !! Appreciate your comment ! Thanks !

6

Thank you for the repsonse !

7

Are people really waiting hundreds of days to get answers, I imagine in that time they would have found answers by other means. Google search is great.

I do have sympathy for any developers in remote regions that lack access to top-tier AI models or expert networks to solve their problems through lack of money or net restrictions. I gather your suggestion is that basically this forum could be the medium through which such developers get access to AI. All be it with volunteer human operators in the middle.

I feel that is likely unworkable.

Do note that Google search AI mode is pretty good at finding answers to Rust questions and finding problems and solutions in Rust snippets you give it or explaining how they work. It's amazing how much Rust code Google search AI mode will write for you if you continue a chat with it. And it's totally free. At least for now, before the whole AI industry collapses.

Also likely for most Rust questions, especially for beginners, an expensive top tier LLM is likely not needed. DeepSeek Harness would be a very cheap option for handling the kind of questions that come up in this forum.

8

Thanks for taking time to understand my POV , may be I am not the expert in all the tech, but I definitely yearn for a workable solution to balance the act of use of AI and content moderation ( which I am in full favor of) , just invigorating a conversation to keep it relevant , and seeking that , we as community are pioneering the solution :slightly_smiling_face: , I appreciate your comment !!

9

Just recent example: I'm a Rust learner, and I figured out, that I can't write Rust code, but I still need to do my job. So, what is a solution? Change profession? Maybe, but I did it simple. I started to delegate all my assignment to AI, and my score gets up. So I needed to write a function returning a real path regardless what case I used to type it. I asked AI and got solution. It worked fine, until a path was a directory. I told AI, it doesn't work for directories. AI corrected the solution and now it works for directories too. I couldn't blame AI, it works perfect, but a bit to verbose for my taste. Only drawback I see, I became too lazy, for example, I need to get str presentation of PathBuf. It's just one call, and even for the call, I ask AI. It gave me 3 variants of the call, I choose only one. I think people shouldn't write code anymore.

10

Depends on how badly they need answers. I asked Nightly archives retention period? 302 days ago. I still don't know the answer, but I haven't tried very hard to find it. I ended up not needing the answer, in the end. I'm sure if it was an answer I really needed I could figure out who to ask, though.

11

This is a community mostly centered around humans helping other humans, by volunteering their expertise. It's not uncommon for the best answers to look beyond the immediate technical question being asked, addressing the larger picture. Or for people learning complex topics, for multiple approaches of thinking about the problem to be presented, reacting to what seems to click for the OP. Beyond "get questions answered", it's a community where people with different tastes and personalities post their takes on the questions, both technical and non-technical, and respond to one another. A social community!

If posting machine-generated answers without putting in much of one's own effort, experience, or personality becomes the norm, the community part of this forum will die off.

In the context of the community helping someone with a technical problem: I for one don't mind reading other people's responses, :heart:ing the ones I consider helpful and insightful, filling in gaps or adding my own perspective, etc. Heck, I enjoy it. But I have little to no interest in reviewing or even reading machine output, especially when no actual human consideration has filtered it or fixed it up. (It's fine to read something like "I asked AI but I don't understand what it's doing here [paste]". It's not fine when all the text presented is from an AI. In those cases, if I ended up reading the post before I realized what it was, I feel scammed.)

From the opposite perspective: If almost all of what people see when they ask questions on this form is AI copied responses, they will soon figure out to skip the middle-human and go ask an AI next time. They'll have zero incentive to stick around and become part of the community. There will be little social interaction because most of what they are reading will be machine generated.

"Know what I mean?" "No. Do you want some more?"

5bt
5bt800×458 30.9 KB

Perhaps that was meant as hyperbole, but I know you're not answering 100 day old posts.[1] The two responses of yours I just saw were to 10 hour and 2 hour old posts.

This is community which is interactive in such a way that threads which don't get activity for a week or so typically don't get any more activity period, and where dormant threads are closed automatically. One result of this is that the same topic can come up and be reiterated multiple times. Which is perfectly fine, as different people learn in different ways, and repeating similar conversations is a perfectly normal human activity. This forum is not like stackoverflow; we're not trying to pinpoint one perfect answer to every question, and to have no repeated questions.

I explain this because it feels like you're trying to assert what should be acceptable here without knowing the forum/community that well. (I could be wrong, but all I have to go on is how you're talking about the forum, your actions so far, and how new your account is.) If you stick around a few months and get to know the community better, perhaps it will be clear how throwing AI responses at questions just to get some answer posted as quick as possible is not a good fit.

None of what I just wrote really addressed how to deal with questions which actually have been dormant for awhile (and by awhile, I don't mean 2 hours, more like days). Actually, I don't think dormancy is the real yardstick here, it's more of a secondary consideration. I think the real question is, should you try to help people out when you don't know the answer well enough to write it up yourself, without significant assistance?

I think it's fine to do so, so long as you actually understand the answer you end up suggesting -- i.e. you went and actually learned the answer well enough on your own -- and you're upfront about what parts you don't know. This is an approach that you can follow without involving AI, and I have done so many times myself. ("I don't know this library but based on the docs, have you tried...."). ("I don't know if it solves your actual use case, but I changed ... and at least it compiles"). I do suggest waiting a bit, like a day say, if it's a topic far outside your domain -- because perhaps someone else in the community who does know the answer, or is more of a domain expert, will reply.[2]

Some problematic aspects of using AI for such answers is that it's just too easy to get an answer to any question, post it without actually understanding it yourself or considering the larger context, and call it a day. Do not do that. If you're going to use AI in this situation,[3]

  • Be upfront and say you asked AI at the top
  • Do understand what it's saying and write it up in your own words before posting
  • Please also apply some critical thought and don't assume it got everything right

AI assistance is fine, if you're upfront about it. AI message deliverer is not. People can go ask AI for uncurated answers on their own. I think most people don't want to see it here (or, frankly, anywhere social).

I like your enthusiasm. But note that even one of the people you were trying to help told you "please no AI responses". Consider taking this as a sign that your exact approach is missing something.[4]

I think my advice is implicit in everything I've wrote so far, but I'll try to itemize it anyway.

  • Before going straight to AI, consider

    • If you can answer someone's question without AI,[5] do it that way. It's a better fit for this community and will usually be appreciated more.

    • Else if the question has adequate answers you don't see a problem with, there's probably no need to supply another via AI

    • Else if the question is far outside your experience, consider waiting a day for others with domain experience to answer first

  • When you do go to AI,
    • Take the time to understand the answer yourself and supply some critical thought
    • Be upfront about your use of AI in your response
    • Write the prose yourself, with your words, your perspective, your personality
    • Never just post AI output. Humans want to talk to humans here, even when getting answers to technical questions. (And it's against the TOS.)

Hopefully that didn't come off too harsh. I care about this community, including its human and social nature. I want others to appreciate those qualities, and I don't want to see them compromised and replaced by excessive machine-generated content. My hope is to get you to see and respect that point of view, and the value of our community. And recognize that the value of our community is more than "question got promptly answered".

I do appreciate that you care about helping others, and that you cared enough to create this topic.

In case it wasn't clear, while I am a regular here, I am not a moderator or on any official teams.

  1. posts automatically close before that point, for one ↩︎

  2. Once you know the community well enough, you may even be able to predict who's likely to answer what questions, and how well. ↩︎

  3. a situation where you needed AI assistance to answer ↩︎

  4. Be aware that posting AI generated answers will almost always be seen as a minimal effort response! After all, if it's pasted straight from the AI output, it probably is minimal effort. (An AI assisted answer that is worth posting, in contrast, will indeed take some effort.) ↩︎

  5. e.g. off the top of your head, or minimal effort looking at some library docs, or posting a citation you already know about ↩︎

12

Many people, including me, want to keep this forum as human-to-human communication. So I think the ToS (rules) reflect that desire. It is not an uncommon rule on other programming forums.

13

Thanks for expressing your views, appreciated !

15

If someone needs to use AI to answer a question, I'm not sure how much they'll be able to verify how good the answer is, and that's arguably the worst way to use AI. But the first question that springs to mind is, why would they answer in the first place if they don't really know? If an AI could answer, I'm sure the person asking the question would check that by themselves. Perhaps they did and weren't satisfied by the answer (which is more likely, seeing the current state of AI).

Also, seeing that the question is already answered may refrain people who actually know the answer from giving a better answer.

I'm not sure I've seen many questions that remained unanswered for long here, so I'd just avoid bringing that AI problem into the help section too.

16

Or worse; an AI answer almost always reads as confident in its accuracy, and you need to ask follow-up questions if you think it's making things up, to get it to tell you what it got wrong; a human expert often writes an unconfident-sounding answer because they're both answering the question as they understand it, and getting you to elaborate a bit so that they can become confident they're answering the right question.

This can lead readers to ignore the expert's answer in favour of the (wrong) answer from AI, because the AI answer tricks all of our human heuristics for "is this an expert's response?". In turn, that's manageable if the AI's answer is obviously wrong, (e.g. "use tokio::task::scope for this" - which is an API that doesn't exist, and can't exist in current Rust because of the scoped task trilemma), but not so much if it's subtly wrong and you need to be an expert to understand why it's wrong (e.g. something asserting that you need SeqCst atomics, when a Mutex would be better).

17

I doubt that people answering on forums such as this one would use "world class LLMs" to reply. I find it much more likely that subpar/free LLMs will be used, which that remote depeloper likely has access to anyway (e.g. who doesn't have access to Gemini)?

If you stretch it you could always make a point that someone, somewhere, might benefit from this, but how rare is that? You also need to weight that against the downsides: how much will be lost by alienating the humans that until now have answered in this forum with their expertise?

My 2 cents are that people posting here are looking for human expertise to help them solve their problem, if that was not the case they could just use an LLM to solve it (or they already tried and that failed). As such I doubt most of them are looking for AI generate answers.

If you have access to a state of the art LLM and want to share it with others you could build a stackoverflow-like website where any question can be answered in this way. After all the reason language-specific websites became popular was because actual people were expert in specific languages. LLMs don't have this limitation, so I don't see why you should follow the same model.

18

Let me just say that if I wanted a machine-generated answer I would have asked the question to ChatGPT or Cloude directly instead of spending the time to translate my thoughts to English (I am not a native speaker) and to write in this forum.

19

Why on Earth do you have that doubt? Seems a strange to idea me.

20

Yes. This is one of my complaints about AI agents in general, and from what I've read, the overconfident tone is something built in because people respond better to it in user testing, and that's more important than the risk of misleading people for the AI companies. It's a big problem. This is manipulation for profit. I find it very frustrating because it could be easily improved if the priorities were different.

Another problem, although it is less frequent, is that the models are always outdated by some number of months (I assume). Some questions asked here are difficult ones, requiring knowledge of current issues, good judgment, and current expertise in the subject area. This forum is incredibly useful for such questions because often subject area experts are here and do respond.

The last thing we need is an AI answer causing confusion in the discussion. Humans are more likely to recognize that they don't know the answer and shouldn't try to respond with outdated or incorrect information.

21

Certainly that was annoying me with whatever versions of ChatGPT I was using a couple of months ago. Not just the over confident style but also the brown nosing going on. Often telling me what great ideas I had and how I was always right.

For a week now I have been using DeepSeek v4 and there is almost none of that. t seems to speak in far more business like manner sticking to the job in hand succinctly. More the direct Finnish style that the US McDonalds server style.

I started to wonder if this is a cultural difference between USA and China that make them train. and condition the things differently.