Roblox AI coding loses its mind?

I don’t necessarily need “help” with anything, but this something I have encountered while coding. And maybe this is a known issue, or maybe not, I don’t know. But Roblox generative AI coding has created swear words and some sort of crisis as a result to my code comment. I’m just posting this as this may need to be looked at.

4 Likes

yet another example if I change the prompt a little bit.
image

3 Likes

the Roblox Ai doesn’t have a mind so it cannot lose a mind because it doesn’t have one
it takes data from other users who have the ai setting thing enabled probably a user created those comments in his script and the ai machine learned those comments and gave it to people

2 Likes

True true, just heavily unexpected for it to start using swear words. Because usually this stuff would be filtered out by Roblox.

2 Likes

not always just like how sometimes the ingame chat filter let users use swear words in the chat and sometimes prevent them from saying normal words like hi or gg

dang the AI is already replacing my ability to schizo comment when things don’t go my way :frowning:

3 Likes

I disabled this a while ago cos it kept adding the n word to all my scripts
image

4 Likes

Can’t ROBLOX just filter the word, I mean how hard is it to filter words :joy:

Hi everyone, we are looking into this.

If there are examples of offensive completions are are getting, please DM them to me for repro steps. There is a safety filter here in place for suggested completions though we’re looking into why it is not triggering in some cases.

Regarding the issue of getting TODOs back in completions: this is distinct from the offensive content issue and as others have correctly pointed out, code models are a reflection of the data they are trained on, and code can at times contain TODOs. We’ll try to see if can improve on that in our future model refreshes.

Thank you for this critical feedback and we’ll try to improve in both of the directions noted above in both our model and safety filter updates. Our goal is to build safe and professional tools you can trust that help you be more productive, and to the extent we’re falling short of that goal, we’ll continue to listen and improve.

1 Like

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.