Code Assist Beta: AI powered code completion

[Dede_4242]

How can people learn how to code from this if it is literally using a lot of bad practices, deprecate code and invest Services and Methods?

Also, you restarted the arguing

[R0bl0x10501050 (RyloRiz)]

@Genichiroblox

Are any examples still triggering repetitions for you?

Just retested, and as far as I can see there are no repetitions anymore. Thank you!

[WillyChan15 (Willy)]

Studio sometimes broke due to the usage of 'alt + ' on my initiative. It shall be a great tool if AI calculation can be improved due to script remaining while not affects the Studio workflow.

[Eituqaru (Eitucaru)]

@Rusi_002

• Alt + \ (Windows)

My gaming laptop does not have a \ button… I can’t activate it.

[Nasa_155]

I also ran into the issue of it using deprecated features, with it using FindPartsInRegion3 instead of GetPartBoundsInBox.

(I prompted it with --Get a table of all parts within or touching LoadRadius distance of Location)

Still pretty cool and still saved me the time of trying to remember/search for what the method name was when I looked up the deprecated one.

[SpeedySanikJr]

I know it’s it Beta but some sort of UI indication that it’s processing would be nice. Maybe something in the output saying “Generation started” and “Generation ended” in the same way Studio will alert you a game is saved so we can actually tell if it’s working.

Sometimes I prompt it and I can’t tell if it’s actually gonna produce code or it’s just dumbfounded.

[SelDraken]

On larger results, I keep getting code that is cut off

[pyxfluff (Pyx)]

Press Alt \ where it cut off to make it continue.

[dduck5tar]

Wow! This is truly incredible.

Mind = blown.

I have never used any of the stuff that has been mentioned in this thread (i.e. Copilot) so this is all new to me. I have to say, it is impressive, and a huge leap toward easily allowing people, from a non-programming background, to have instant access to coding concept and methods, without swapping over to a search engine to find a decent example to learn from.

This truly does change everything, just as AI is already doing in virtually every sector in which it finds a home. Hat’s off to the idea-smiths, the coders, and the brains that can even think this, never mind actually achieve it!

As this tool develops and the vocabulary and diction progresses, it does however, foretell the beginning of the end of the DevForum. I mean, why ask people stuff and wait days for a response, or get none at all. We can now just type a correctly phrased prompt into Studio and viola! An instant answer, without flowery response or deflecting personal opinion.

Lua is a joy to use and implement, and as soon as Roblox improves the graphical capabilities and procedural power of this engine, Unreal and Unity will be looking very nervously over their shoulders…

[MoonRocketApollo]

@C_Corpze great questions, we’re discussing now & hope to have an update on at least some of these ideas as soon as possible.

[MoonRocketApollo]

Hey @SelDraken the team is digging into this now – long responses are sometimes truncated. For now the suggestion is to go to the end, remove any partial lines of code and either use the keyboard shortcut or try something like

-- complete the code

Thanks!

[MoonRocketApollo]

@SpeedySanikJr

ome sort of UI indication that it’s processing would be nice. Maybe something in the output saying “Generation started” and “Generation ended” in the same way Studio will alert you a game is saved so we can actuall

We’re working on this now and will have an update that better indicates when the model is working and to better indicate when there’s no response (and as much as we can, why there’s no response so you can take action).

Thanks!!

[qscythee (scythe)]

Code assist is pretty cool so far, however, lots of room for improvement! Considering it’s a beta, not too bad at all. I haven’t found myself using it in production code but do I like to give it challenges and conversations. Speaking of, here are some funny convos:

[Malte0621]

Roblox it self is still in beta lol. Sorry if its not directly related but I just had to mention it.

But tbh, its cool and all. But it needs more training and it needs to be more accurate with less junk (filtering out useless/repeated code).

[Malte0621]

Seems like it just broke down for me. Its saying the following in the studio log files:

2023-03-21T15:34:37.034Z,22.034773,5678,6 [FLog::Output] Info: RobloxScriptDoc::open - start
2023-03-21T15:34:37.080Z,22.080765,5678,6 [FLog::Output] Info: RobloxScriptDoc::activate - start
2023-03-21T15:34:42.449Z,27.449430,5678,6 [FLog::Output] sendMLCodeCompletionHttpRequest ParseJson Error : InvalidJsonDocument
2023-03-21T15:34:45.914Z,30.914095,5678,6 [FLog::Output] sendMLCodeCompletionHttpRequest ParseJson Error : InvalidJsonDocument
2023-03-21T15:34:49.855Z,34.855522,5678,6 [FLog::Output] Info: RobloxScriptDoc::activate - start
2023-03-21T15:35:11.586Z,56.586761,5678,6 [FLog::Output] sendMLCodeCompletionHttpRequest ParseJson Error : InvalidJsonDocument
2023-03-21T15:35:46.751Z,91.751350,5678,6 [FLog::Output] RobloxMainWindow::requestDocClose doc.docType=2
2023-03-21T15:35:46.752Z,91.752350,5678,6 [FLog::Output] RobloxMainWindow::forceDocClose doc.docType()=2
2023-03-21T15:35:46.752Z,91.752350,5678,6 [FLog::Output] RobloxDocManager::removeDoc type 2
2023-03-21T15:35:46.764Z,91.764359,5678,6 [FLog::Output] Info: RobloxIDEDoc::activate - start
2023-03-21T15:35:46.787Z,91.787384,5678,6 [FLog::Output] RobloxBasicDoc::closeDocument 000001EF57574110

Notice the sendMLCodeCompletionHttpRequest ParseJson Error : InvalidJsonDocument

Is this some kind of backend error from roblox or is it something locally for me?
Some help would be appreciated.

If needed, I could do a network scan and see what the “Invalid json” actually is.

Edit: Seems like the server is returning HTTP 204 (No content).

[B1GBRAIN2 (WUT)]

i was hoping such a feature got released like 2 yrs ago anyways good thing it did got released

[SpaghettiANDchees]

I think this will be quite helpful, especially when your stuck in a situation and don’t quite know what to type.

[Malte0621]

Can’t we just change "max_tokens":500,"temperature":0 to have different results for the generation? I can see that the AI is told to be 100% logical. Maybe it would be fun to change it?

Anyhow, maybe the "stop":"\n\n" is the cause of the problem? Or it might just be something else on the backend .-. (Edit, its most definitely something else on the backend.)

Even making the request manually causes a 204

(also, unrelated, its my birthday today )

[sw4gswagsw4g]

I feel the same about this AI assistant as I do about the usage of GPT4 for simple apps. Seems cool for one off, isolated situations, but the moment I need it to tie into my object-oriented architecture it becomes useless.

If I have to direct it to call specific helper methods from my custom objects I’m already better off writing it myself.

Eventually I’m sure these models could be trained against an existing code base but until then I don’t think it would help my productivity.

[killeralfi]

I haven’t even seen this generate a suggestion once since I enabled it.