I might do it, but I need to be absolutely certain that the forum moderators have no connection to the people who review the edits on the OpenWrt wiki. I'm not willing to waste another minute of my time. Can you imagine doing the work only to have it thrown in the trash?
When I've followed precise instructions on this very technical forum, they haven't always worked. I've had to accept the risk that something might go wrong, and I don't blame anyone when things did. I did it because I wanted to, accepting the risk.
Don't forget, curiosity moves the world. Without curiosity, there is no knowledge.
And I'm very curious, more than a cat. 
Yeah, this happens from time to time. Sometimes it's as little as a human induced typo, other times it is a user who doesn't know exactly what they are doing (or made a mistake), and it can even be related to the fact that syntax and methods do change over time as OpenWrt evolves.
But, it is safe to say the accepted solutions are often correct.
That is fair and reasonable. In the case of AI specifically, when it is your own device/network and your own (intentional) use of AI, the risk is yours and yours alone. Sharing that AI guidance is another matter altogether.
The reason for the rule is there to prevent the proliferation of incorrect and even damaging/dangerous guidance caused by AI. For example, a novice user skimming the forum might see a post with AI generated config code and may implement those changes without realizing that the config is flawed. Beyond that, AI will crawl the forums and find that information, assuming it to be correct and authoritative. That leads to a feedback loop of incorrect information making AI worse at its guidance and polluting the forum with posts that have information that is even more off the mark.
Curiosity is great.
The problem with AI is that you often need to be a subject matter expert to understand if the answer is correct (in part and/or in whole, and this goes for any/all subjects, not just OpenWrt/networking). And most people who use AI are not SMEs in the fields for which they ask AI to help (myself included)... if they were already experts, they wouldn't need AI, after all.
Many (but not all) of the AI posts here include configs or scripts which are expected to "just work" but contain no explanation about what the code will actually do (line by line) -- this doesn't help anyone learn (especially if blindly executed), but it does expose users to the dangers of AI hallucinations causing damage to their configs/networks and even bricking devices.
Beyond that, if AI is believed to be authoritative, why post the AI solutions here rather than just have each person ask AI for themselves? And why have a forum where actual humans help solve problems?
2 Likes
I can see the wall is going to be quite high to climb.
No offense, but when someone makes a lot of excuses, I tend to be suspicious.
I clearly stated that I personally tested it on four of my routers. I'm not "retweeting" anything from the AI.
TIA
We have seen AI fail miserably. For example, a user once posted a table comparing specifications (chipset and other hardware features) of several devices when asking about the supportability of a given device/family in OpenWrt. Seems like a pretty silly, but seemingly simple thing for AI to do. But, as it turns out, it was entirely wrong, despite the fact that simple (manual) searches o the web could find teardowns and other sources with the correct information.
Another thing we see frequently... AI generated shell scripts -- they often start with the shebang:
#!/bin/sh
The problem with this is that OpenWrt doesn't use sh for its default shell. It uses ash. There are differences in the features and syntax of ash vs sh when you're writing more complex scripts, and yes, this can really mess things up when a script fails to execute properly.
[EDIT: As pointed out in a PM, the shebang is actually correct and appropriate for OpenWrt, so this was a bad example. There are certainly other things we see in scripts that may lead to problematic outcomes, but this is not one of them.]
I don't think it's worth continuing with this topic.
You win.
Quite frankly, just to add to the above.
The typical ai generated content is very easy to spot (the 'perfect' formatting, with many subheaders, and graphical symbols, alone, before even thinking about mdash and friends), beating around the bush with much prose, but little substance. It is not easy on the eyes, especially to experienced users who might otherwise be inclined to help you.
Talking about myself only, once you invoke the magic AI gods, I'll quit the discussion - why:
- I am not paid to teach the AI
- the resulting text is long and non-concise, with really annoying formatting
- I can never be sure which of the sentences (correctly) reflect the (prompt-) 'author' or are just AI slop, in many cases a user lazy enough to use AI, will not properly read the results (the questions they expect others to answer) themselves, so why should I spend my time on this again (the 'author' didn't care enough either)
- AI generated configs are much harder to proofread, they may lull you into thinking that large parts are just 'defaults' (so you skip over them too quickly and missing the subtle issue), they are very prone to get important syntax details wrong (quotation marks,
"vs'matters, before even thinking about backticks or UNICODE specific variants of speech marks, inverted commas, etc.) - very hard to spot - AI generated configs often differ quite significantly from 'normal' ones, even when technically correct, they look off, making it much harder to spot the problem
- and yes, AI isn't intelligent, it doesn't know what's right or wrong, it's not sentinent, it just combines snippets found elsewhere based on probabilities. it happily 'learns' incorrect input, which is likely to be found on technical forums - because users show their incorrect configs all the time (after all, getting help doesn't work otherwise)
- it has no idea what's wrong in the original config, nor what eventually fixed it down thread
- it will fail badly whenever there are are changes in accepted configs (e.g. swconfig vs DSA), both in the sense of suggesting swconfig format for DSA device (which just migrated in the last OpenWrt version) or suggesting DSA for swconfig devices - or happily mixing both concepts (which doesn't work). It will be overwhelmed with x86-like devices, which have no switches but independent network cards.
- AI bots are actively harmful to the forum, wiki, git servers used by OpenWrt, constantly being engaged in -effectively- DDOS attacks against server maintainers.
this behaviour proves beyond reasonable doubt that contemporary AI is not intelligent, but blindly follows every (generated) link, rather than understanding the common web frameworks (clone the git repo, do your learning offline - don't hammer the gitweb interface!!!)
So if a user is too lazy to write their own text, that isn't respecting those they expect to get help from. I have no motivation to reward that, nor to teach the AI behind the slop.
As a moderator, you see a lot of really bad AI generated content - things many of the more casual forum users won't even get to see.
10 Likes