GPT to answer questions?

[spk100]

Has any one implemented a GPT to answer user questions?

If yes, how did you do it?

My idea is to implement an AI APi that will answer user queries after x number of days if it has not received any replies from any users.

 

[Mr Lucky]

Has any one implemented a GPT to answer user questions?

I tried it for a while, there was an addon that is now gone.

Overall members were not in favour. Answers were often wrong or stupid. Some members just asked stupid questions to bait the bot.

 

[BassMan]

I can agree with Mr Lucky here. I tried it on two forums with different niche, but there was no benefit from it. The bot is retired for quite some time now and I'm not thinking to wake him up again.

 

[spk100]

Oh.. I did not see the add-on before. I was thinking of training it with some data and then asking it answer questions, just like how a custom GPT would do.

 

[CedricV]

Oh.. I did not see the add-on before. I was thinking of training it with some data and then asking it answer questions, just like how a custom GPT would do.

That’s the way to do it. The default non-trained GPT is trash.

 

[spk100]

That’s the way to do it. The default non-trained GPT is trash.

True. Just wondering if there is a good way to go about this.

 

[Alpha1]

It seems to me that this has massive opportunities if done correctly. It can easily become a mass garbage production script.

I remember the auto tagger script on vbulletin creating a mess that took years to repair. This could be so much worse if we allow it to mass post freely.

Forcing it to reference all statements with citations, forum quotes and links could be one way to limit garbage answers.

It should also learn and adhere to the forum rules.

I think a first version would require checking the answers manually. LLM s tend to be verbose and elaborate, so manually checking it takes a lot of time. It's easy to assume that the text it spits out is correct. Often it turns out that is not.

 

[spk100]

It seems to me that this has massive opportunities if done correctly. It can easily become a mass garbage production script.

I remember the auto tagger script on vbulletin creating a mess that took years to repair. This could be so much worse if we allow it to mass post freely.

Forcing it to reference all statements with citations, forum quotes and links could be one way to limit garbage answers.

It should also learn and adhere to the forum rules.

I think a first version would require checking the answers manually. LLM s tend to be verbose and elaborate, so manually checking it takes a lot of time. It's easy to assume that the text it spits out is correct. Often it turns out that is not.

That's true. My idea was not to auto post stuff and instead cross reference old post content, public sources and then post a relevant response with an option to moderate the answers (just like a new member).

Currently my forum has moderation set on all new member until they reach x posts. Maybe this can also follow that.

 

[Alvin63]

Not for answering questions, but it's very good in the AI Content Moderator programme by Genesis.

I ready something in the media recently that some bigger site's chat bots were talking inappropriate stuff (If asked presumably!).

 

[Alpha1]

Other things that an LLM could do for Question threads:

1. Propose a question as the thread Title. More often than not people post a title that is not a question and then there is no question to answer.
2. Propose a 'Solution / Best Answer' or cast a vote for it. We have too many question threads to do it all manually.
3. Properly tag it.

 

[spk100]

Other things that an LLM could do for Question threads:

1. Propose a question as the thread Title. More often than not people post a title that is not a question and then there is no question to answer.
2. Propose a 'Solution / Best Answer' or cast a vote for it. We have too many question threads to do it all manually.
3. Properly tag it.

Great ideas.