LLM based Replies to community posts

[Kirankumarbs]

Hello Everyone,

I’ve been in the community for a few weeks and really enjoy reading the problems and production-focused solutions shared here.

Recently, I’ve noticed some replies(solutions) that look quite verbose and seem like they might be generated by LLMs or AI tools.

Please don’t take this the wrong way, but I personally find this less helpful. I usually try these tools myself before coming to the community, so seeing similar responses here takes away a bit from the value of human, experience-based discussions.

Maybe I’m misunderstanding the expectations here, so please feel free to let me know if AI-generated replies are considered okay in the community.

[szymon_dybczak]

Hi  @Kirankumarbs ,

I agree with you. The sad part is many of those answer are simply wrong. If you want to use LLM that's fine, but we should always validate if the suggested answer works. Unfortunately, in most cases the people don't put any effort and just blindly copy/paste replies generated by chats (which by the way are mostly wrong when it comes to newer functionalities in Databricks).

[Kirankumarbs]

This is something I’ve noticed as well.

Many replies list broad options or multiple solutions instead of giving a focused answer. As I mentioned, I usually ask an LLM before coming here, so when I see similar exhaustive responses, it makes me wonder what the purpose of a human-driven community discussion is.

I do think it makes sense to use AI for research and then combine that with your own production experience when replying. That adds real value  rather than just pasting a large AI-generated answer here.

[MandyR]

Hey guys, I appreciate you bringing this up. We do have an AI usage policy in our Code of Conduct that I recently asked Legal to update:

Use of AI Tools: AI tools may be used to assist with formatting, clarity, and organization, or to draft ideas that are fully verified by the poster, but automated posts (e.g., via bots) are not allowed in the forums. AI-generated solutions must never be shared without human verification in the Databricks Community forums. Users must disclose when AI-assisted content is included. Repeated misuse of AI-generated content may result in content removal or account restrictions.

That said, it's basically the honor system because I don't have a way of confirming that any verification was done. 

Some people have been adamant that as long as posts are disclosed as AI-assisted, and they say they are verifying them it should be enough and that in fact, most people in these forums are quite happy with this.

Others say that AI ruins the spirit of collaboration in the forums, and makes it seem impersonal and less valuable. 

What do you guys think? I would really like to know what's best for our customers so I can steer the direction of the policies in our forums. It's a weird place to be in right now. 

[Kirankumarbs]

@MandyR 

I really appreciate you following up on this topic. It shows how dedicated the community organizers are.

Sharing my personal perspective as a Staff Engineer who already uses tools like Claude, Cursor, and Copilot in day-to-day development.

For me, when I run into an issue, my first stop is usually these LLM tools. I come to the community mainly to learn what experts or other users have actually experienced in production, rather than reading theoretical solutions.

There are already plenty of theoretical answers available through Google or AI tools. What I’m really looking for here is first-hand experience, because I’d rather try proven approaches than experiment with five different options at work.

If someone doesn’t know the answer to a community question, it’s probably not helpful to just copy a response from an AI tool and paste it here. Any developer can already do that themselves.

In short, what I (and likely many others) value most here is interactive, human-driven discussion. People sharing their experiences, lessons learned, or insights. Sometimes someone may have already faced the same issue, or a Databricks expert may know the right direction internally. That kind of input is what makes a community truly valuable.

[Kirankumarbs]

For example, replies like this aren’t really necessary:
https://community.databricks.com/t5/community-articles/ci-cd-on-databricks-with-asset-bundles-dabs-a...

If you look at the original post, it already shows a practical approach and clearly reflects the author's real work and experience. Replies like this can give the impression of spam to the reader.

Instead, I would expect responses where someone shares their own experience, lessons learned, or practical insights related to the topic.

Another example: https://community.databricks.com/t5/data-engineering/databricks-spark-ui-showing-1-executors/m-p/150...
Louis Frolio’s reply was good and genuinely helpful to the author. However, the other agent-like reply gives the impression of spamming and doesn’t add much value to the discussion.

Sorry if this comes across as complaining. I’m just sharing my opinion in the hope of keeping the community standards high and avoiding unnecessary spam.