News:

SMF 2.1.4 has been released! Take it for a spin! Read more.

Main Menu

How is everyone implementing AI into their SMF forums?

Started by Biology Forums, June 04, 2024, 11:23:55 AM

Previous topic - Next topic

Biology Forums

I'm sure this idea has been discussed extensively already, but I'm curious how members of SMF have used AI (such as ChatGPT) already to help draw new activity and produce a better user experience.

Arantor

There's a limit to how much people actually are/aren't. The general trend seems to be 'not really' but one could argue that's reflective of the general disinterest in changing the forum format too hard.

I've seen some support forums (e.g. Laracasts for Laravel) attempt to offer a first-pass attempt at an answer by feeding your question in with some context about Laravel, and it tries to give you a reasonable attempt at an answer. My experience with this has been hit or miss so far, but of course this is a given: AI does not *understand*. There's no comprehension there, it just produces words based on what's statistically aligned. If your problem is common and previously talked about, fair chance you might get an answer but this is not without peril.

For example, there was a series of posts recently on social media demonstrating this at scale. People would ask for the river crossing problem (farmer + fox + chicken + corn need to cross a river, farmer can't leave fox + chicken alone or chicken + corn alone but can only cross with one at a time), and get the stock answer. But the minute you change it up such that it still *resembles* the puzzle but fundamentally changes it, it still tries to hammer the solution through. E.g. people would ask it about a farmer and a chicken crossing a river, on a boat called Wolf. Considering it's just a farmer and a chicken, it's one crossing, but this usually ended up as 5-7 steps.

As for other options for AI, I've seen people use it to offer a sort of general chatbot, but this inevitably leads to a place of abusing it for comedy purposes rather than any tangible benefits. Also integrating out to Midjourney or DALL-E for image generation I've seen done (I've even seen one place use DALL-E to do avatar generation which I thought was interesting but it was very haphazard what it generated)

Probably more interestingly, I've seen people use it to suggest better titles for topics (as a one line summary of the opening post rather than the usual things people do), I've seen people use it for summarising longer topics, and I've seen people try to use it for spam detection. All of these with varied levels of success, but there's some potential there, it's just whether the physical cost of implementation (since ChatGPT APIs etc. aren't free) becomes worth the perceived benefits.

Advertisement: