Hey there! A while ago I made the website https://60secondscripting.com which contains GPT-3 generated articles for simple Roblox Lua basics. These are all generated with a prompt to GPT3.
Today though I wanted to run an experiment in how well this type of AI can address arbitrary questions about Roblox Lua. The best way I thought of to do this is pulling scripting support posts from the devforum, and having the AI attempt to answer them. I gave it the name ‘Scripty’, and a fun little icon
The AI doesn’t and won’t ever actually post on the devforum, I assume this goes against the rules, but even if it didn’t it would just be clutter - these responses aren’t always perfect.
Just add the post ID, found in the address bar to the end to try and find a post. Only new posts from today onwards are being processed, and posts over around 1,000 characters aren’t processed because of token limits.
I’m also interested to know how people on here feel about this. Do you want your questions out on a website like this, and are you ok with them being answered by an AI? I’m happy to implement some kind of system where you need to submit an article. In the meantime, you can email me @ luke@rtrack.live if you want your post removed (or for anything else related).
Please let me know what you think!
Potential in future
I’d like to train a model on devforum replies to improve it’s accuracy and knowledge about new features, I might do this in future. This depends on how the community feels about something like this, and of course ToS
I would strongly suggest you do not do this without consulting Developer Relations first, or at least obtaining user consent. This is pretty murky water as far as what’s permissible.
There is already heavy controversy over how training data for AI is scraped from the internet, specifically around consent and rights. Nobody on the DevForum has consented to their posts being fed into machine learning algorithms of any sort thus far, and without that consent, it’s unclear and not generally understood whether what you are doing here is even legal.
I don’t say this to scare you, but I say it because it’d be completely irresponsible to just do it and ask for forgiveness later, whatever your own personal opinion on such practice is. You should talk to the people in charge here about it, and see what their perspective is on third parties explicitly using the responses of users here as training data for generative models.
I definitely won’t make the model without more feedback on it - this version is just reading the question and running it through GPT-3. Scripting Support is also open to the public to read, so it’s not putting anything out there that isn’t already public. But thanks for the other stuff, as I said in the post looking for feedback on how people feel about this, of course if anyone at devrel etc has issues with this I’ll happily close it down, this was just an evening project!
I understand that everything we do on roblox as developers, including the forum stuff, all belongs to roblox. That’s why roblox deletes creations, locks accounts, restricts users with just an explanation or automatic answers.
So really the one OP should ask for permission is roblox. But large corporations don’t listen to individual requests and are guided more by the possible harms or benefits of something, and that’s when they take an action.
In the case of AI, at least so far, it has proven to be very useful and of course there are people who have been affected. But the solution is not necessarily to stop doing AI. In time it will be known whether the project is more beneficial than harmful or vice versa, and roblox will take an action such as supporting or absorbing the project or fixing the flaws or simply shutting it down. For the moment nothing is known and nothing will be known if it is not continued.
This will be heavily abused, reminds of me a person who was doing the same then got suspended. Plus scrapping data (or just crawling) off discourse sites and roblox is illegal.
I agree, I still remember The Tread, (“People That Use ChatGPT To ‘Help’ (Bring Popcorn It’s Getting Caothic))
And it all started because of the guy which sued ChatGPT to answer…
Thsi might have a similar effect
Scripty is now given a summary of the question asked in the devforum scripting support post, instead of the question itself. I’m hoping this will improve response quality.
Source?
If this is true, Google is technically breaking the law, as discourse forum pages are easily searchable via Google, and that requires some kind of web-crawling and data-scraping.
The real problem IMO is if AI-generated blog pages become profitable to the point where human-made blogs are left dead in the water, and only AI-mass-produced blogs survive with decades old regurgitated information.
