Skip to main content

I’m curious what we can do when a Suggested Answer to a question is incorrect in order to prevent users from getting the wrong information. For example, here the rep asked “Can you send hyperlinks in SMS messages?”

Suggested Answers stated, yes, you can, however this is not the case, and we cannot send hyperlinks in SMS messages. 

Is there a way to correct this? 

On the flipside, if a user marks a card has helpful that has incorrect information, is there a way we can correct this. For example, the suggested answer shown here is incorrect. 

Edit: After posting this, I did notice that the answer changed once the card was flagged as shown here, which may render the first part of this question unnecessary. 

Another edit (that may help people with the same question): I played around with the verbiage on the card that was marked helpful but had an incorrect answer. I also marked it as unhelpful.

These changes updated the suggested answer to something that was correct. 

All that said, it seems like simply playing around with the card and the 👍 👎 can eventually get you to where you want to be! 

The best way right now is to manipulate your Guru content, ie changing or adding information in the source content in a way that will change the AI’s answer. I’ve had spotty luck with this method though, sometimes it changes the AI’s answer and sometimes it doesn’t. For example I recently changed info on a card to fix a question that was giving the wrong answer. i made it clear in the card what the answer should be, and then i tried asking the question again. it seemed to only half work - for some reason adding or removing the question mark in the prompt would give me a different answer (exact same question otherwise). 

 

So retroactively establishing an answer is clearly a challenge to solve and it’s for probably a number of reasons, like the limited usage of LLM for the AI to reference for future queries (this is not to say i dont support this decision but it is a factor), and even the wide variety of ways a user base could input a query prompt (even just punctuation, apparently)

 

is there anything we can do to improve the answers to questions aside from updating existing content? im not sure, but i do believe an important step is to ensure the AI is only pulling data from the guru environment and not outside of it. From my experience so far, i dont think it’s advantageous for the AI to give answers that are sourced externally.. i plan on making a guru post about this topic soon to give more context on my thoughts around this. 

@timsubealdea  Thank you for your response! I discovered the same after posting my question (will do more researching before posting next time), and found the same result you did. It seems like tweaking your content, and/or whether the card is helpful can eventually get you to a more appropriate Suggested Answer.

I have since updated my original post with these edits. 

Reply


Terms & ConditionsCookie settings