Skip to main content
/

Site Navigation

Your Account

Choose Language

Discussion Topic

Possible AI answer bot?

@krisrodriguez - Can you fire off an email as this person is using AI answers - Stuey

Reply to discussion Subscribe to discussion

Is this a worthwhile discussion?

Score 4
Add a comment

8 Replies

My take is probably, but likely yes. Based on this answer being the same as so many, it looks and smells like AI: https://www.ifixit.com/Answers/View/8818...

Was this reply helpful?

Score 3

4 Comments:

@nick, yea, this is definitely being generated by AI. But, would it still be helpful to OP if they followed the steps?

by

@krisrodriguez I need to review a sample of questions. I just checked two to get a quick answer.

by

@krisrodriguez Example 1: https://www.ifixit.com/Answers/View/8859... - Is the fuel pump easily accessible, or does removal require you to remove the gas tank? Foreign cars usually have a door, but the OP did not check. I would consider this unusable, especially for someone with this major misstep. I know it's probably under a door and would be able to cope, but does the OP? NOT EVERYONE KNOWS THIS OR HOW TO CHECK QUICKLY. I think this one needs to go, with caution to the OP about tailoring the responses.

Example 2: https://www.ifixit.com/Answers/View/8854... - usable, but it's a straight copy which isn't tailored to the OP's question.

Example 3: https://www.ifixit.com/Answers/View/8860... - same as 2, usable but posted as-is.

by

@nick, thanks for looking into this.

by

Add a comment
Most Helpful Answer

I'm a little torn. The answers do seem AI generated, but the answers also seem on topic. I would be more inclined to instantly ban them if they were copy-pasting the same thing across multiple questions. @danj j and @nick , in Stuey's answers, are they at all helpful in assisting OP solve their problem?

Was this reply helpful?

Score 4

11 Comments:

Hi Kris, it’s more of spraying a lot of things than a focused answer. Can you just email him to refrain from deviating from the question and limit the AI involvement.

by

Hi @krisrodriguez

I've already posted a comment to the user regarding using AI in an answer just to inform them about their responsibilities to ensure that their answers are relevant and accurate and not just what the OP of the question that they responded to in this way could find by searching online themselves.

This was a few days ago and there was no feedback from them, but I noted that their answers have improved slightly away from just what I believe to be AI copy and paste.

Time will tell.

Cheers

by

@jayeff I think we should restrict AI to being only allowed when amended by the person posting it given it's GIGO when misused. I know how to use it well, but the majority put a bad prompt in and get garbage out.

by

@nick

Agreed.

If they "personalize" it at least then they have made the effort to read it through and who knows, may even delete some of the more irrelevant information, rather than just copy and paste

Cheers

by

@jayeff There's coruses online for it that are literally low cost or free to actually make it useful. I'm not saying mandate an AI certificate or something like Google Cloud, but don't use the response as-is. People often prompt it with absolute (for lack of a better word) dogshit or use the original poorly written question from the OP and get bad responses, and you end up with garbage and then post it as if it's viable. No, read the post, tune the response.

AI is not mature enough. Yes, I can use AI well but it's a learning process and I can't exactly teach everyone who makes an account about how to not write hot garbage. Even if I wrote a guide to effective prompting, it would probably be ignored because GIGO is faster then effective prompting.

by

Show 6 more comments

Add a comment

To futher add to the discussion on why I cannot stand AI on a forum like this let me take this question as an example:

ORIGINAL POST: https://www.ifixit.com/Answers/View/8853....

This is the base response I get from ChatGPT:

Now let me show you what it says, both with effective prompting, AND crap prompting:

Now, when I reload ChatGPT with the same base question to remove any history, this is the same response with the original prompt (the answers post):

When I prompt it with the bad foundation with the model (L3150):

Now, when I prompt it with my initial answer (again, simulating bad use of AI like so many AI Answers post use):

When I prompt it with an effective iterative prompt:

As you see here, it can be an excellent way to get some points quickly, but it cannot be used as-is on your first go - it requires iteration and human verification to make sure the information is correct. The only reason I got an effective answer is that I could feed it a bad post, a good answer, and properly iterated the prompts. At this point we've beaten the horse to death on how AI is nearly useless with bad use, but I am using this to prominently highlight why I want to see a standard such as human corrections being a must if you want to keep your AI posts up. If the discussion is left at "check your work" maybe someone on the iFixit side can make a guide on how to effectively prompt AI to make these answers bearable, as a last ditch effort. Take note it even missed my point about the $20 meter being sufficient for LV work and assumed my Southwire was $20 - it was closer to $40-50, so double the price. I left that in on purpose to show the hallucination problem.

***This is the current version of what I have: https://www.lowes.com/pd/Klein-Tools-Dig...

Was this reply helpful?

Score 2
Add a comment

@krisrodriguez - Not to add salt to the wound here’s another run of quick AI answers within this users answers Jakub Topolski

Was this reply helpful?

Score 1

3 Comments:

Haha. @danj, thanks for being so considerate. But, these are good catches. I've gone ahead and done the same for this account: temporary ban and will reach out with an email explaining why. Part of me wants to hope that they are just trying to help, but the truth is that these incorrect answers can make an already bad repair problem, much worse.

by

@krisrodriguez - Mod’s can only go so far with cases like these. It’s your @ifixit that makes the difference. I’ve already removed a few way off answers. But until the person gets a notice we are just chasing the problem not dealing with it.


As always Thanks!

by

@dan, I agree. I think this will be the way to go when we come across a user who is just copy/pasting AI answers.

by

Add a comment

Hey @krisrodriguez I suppose I found another one: https://de.ifixit.com/Antworten/Ansehen/...

Answers are not wrong but too general and less detailled for this topic.

Was this reply helpful?

Score 1

2 Comments:

I answered it myself and made it a moderation move 😎

by

Add a comment

@krisrodriguez Not necessarily bad answers, but certainly AI: 4658802

Was this reply helpful?

Score 1

1 Comment:

Yep, title and ending gives it away. *facedesk* - edit your AI responses, people.

by

Add a comment

@krisrodriguez we have had a discussion about the use of AI (KI for our European colleagues :-). It is my understanding that it is up to us to moderate this. We should remember that we all decided on an iFixit Community AI Policy "Caution is essential when using AI for repairs. AI lacks hands-on experience, real-time assessment, and may provide inaccurate advice, risking worsening issues. Its static knowledge may not address evolving technologies or unique circumstances, potentially leading to unsafe situations. Balancing AI with human expertise is crucial for accurate and safe repairs" and "If a community member uses AI to write content for iFixit, they must check for accuracy and safety 100% of the time."

To me, this means that it is not a @krisrodriguez issue but a moderation issue. We should adopt approaches just like @dadibrokeit did on here I have always advocated that the OP of the AI/KI answer must be transparent and openly make a statement declaring any AI/KI answer as such.

Let's moderate according to established guideline. To me this means Klappe zu, Affe tot ;-) when it comes to AI/KI.

Was this reply helpful?

Score 4

11 Comments:

The issue is a combination of education (poor use) and nondisclosure for me :/. It's tolerable with a good base question, but it's still a crapshoot. I've tested it with a properly abysmal sample question I rigged up in 5 minutes (yes, it was unusable), then revised the question with a fresh ChatGPT session. Thw refined one came out better, but it still left way, way more then I'd like to see.

I didn't put an AI disclosure on this tester, but I did note it was for a sample. I didn't want to give too much away why... See here: https://www.ifixit.com/Answers/View/8870...

There's no way to remove the AI markers, no matter how much I can think of to clean it up. I would probably call that a good thing, as the amount of editing I'd need to do to remove it would make it such you're better off skipping it.

by

@oldturkey03 - if it was just one or two posts then I have no problem just dealing with them. The issue I pressed was when someone does a good dozen or more which spans many products. Having Kris pen a warning note to the poster helps keeps the mess to a more tolerable level.

by

@danj this assume that this person even uses their email account. If we comment on this, the OP of the AI/KI answer at least sees that we are serious about this. You as a mod have always the opportunity to delete the account, which gives the Admins a chance to review. If nothing else, you can start to downvote with the rest of the mods following. That will alert those that read the answer that there is something fishy going on. If we want to continue to run answers the way we think it would be best, then we need to take the initiative. I have a feeling the Admins are busy with something else. You've all noticed how quiet things have gotten......

by

@oldturkey03 A lot of these AI Answers people pump them out without regard for quality. I doubt they give a %#*@ about trying. I hate to say this, but we need a way to tag it as an AI answer the person who posted the AI answer cannot remove. The ONLY WAY to remove it aught to be admin stepping in or deleting the answer and posting an original.

Do you think I wanted to use the nuclear solution? No, I wanted to give them time to prove us wrong but it keeps happening.

by

Thanks for the reminder Mike, after thinking twice it might really be the better way to moderate the "answers"and show the writer (and the public) that we are serious about AI. I will try to use that approach.

by

Show 6 more comments

Add a comment

Perhaps we should consistently mark all the answers that catch our eye in the first line like this:

"Warning: This answer is generated by AI. Use it with caution."

Was this reply helpful?

Score 3

1 Comment:

This might be a good way to get ahead of these. Let me talk with the team internally to see what they think. I'll follow-up soon.

by

Add a comment

Join the discussion

Dan will be eternally grateful.