AI Assistant code completion does almost nothing. Normal?

Answered

I have been using AI Assistant since it was publicly released a couple of weeks ago and I love some of the features. I had been using Copilot for about 3 months prior and I like the way AI Assistant is more deeply integrated with the IDE. But one thing has been pretty disappointing: code completion/suggestion. Since I started using it a couple of weeks ago it has only suggested something a handful of times. I'm not sure if this is normal or there's something broken. In contrast, Copilot was constantly suggesting things for almost every line of code, and they were useful suggestions.

I've used AI Assistant with a variety of projects and file types (Laravel, WordPress, Shopify, PHP, JS, CSS, Liquid, etc). And I'm not seeing any IDE errors. Is this expected behavior with AI Assistant or is there something wrong?

The other AI features (chat, documentation, etc) are great, but the code completion is practically useless this way.

4
7 comments

I totally agree with your opinion. I keep trying to find out what benefits it can give me, but I can't find anything that it can help me with. AI does not fully understand the essence of the code and offers not fully thought out refactoring. Always limited code analysis.
I didn't find a single function that would help me write code faster, which I already know how to write by my self. So what can I say, maybe it is suitable for a manager who is starting to learn programming, or the student that he finds it difficult to program fundamentally?!

0

Also finding myself in a similar predicament, unfortunately. I was hoping that assistant would speed things up by non-invasively suggesting auto-complete by inferring correct pattern from the context. But it seems to suggest incorrectly too often. So at the moment, it seems like more of a mental noise rather than a time saver :-/ 

Regarding a comment that it might useful to a student, I'd even venture to say that this feature might be pretty harmful to the beginner developers – without careful scrutiny of each of AI suggestion may result in riddling the code with subtle bugs. 

One cool feature that seems to work great is documentation – generated content is on point and useful (at least in my limited experience)

Commit messages are too verbose and lack substance in most cases, so not there yet.

0

Admbox1 I totally agree about the documentation and commit messages. I use the generated documentation all the time with very few changes. But I have never used a generated commit message because they are just so darn long. Plus they seem a little pointless because they basically just summarize the diff in paragraph form (i.e. listing what files changed and what changed). I prefer commit messages that summarize what the goal was or what problem was fixed. You can look in the git diff to see what changed.

Re: inline suggestions – after my original post, I found out that I have to enable automatic inline suggestions per project. I thought this was a global setting. So I'm assuming that's why I haven't been getting very many suggestions. We'll see how the next few weeks go. I had been using Copilot before this and loved it. I definitely had to edit or totally ignore the suggestions pretty often, but overall it felt seamless and very helpful IMO. I'm hoping AI Assisstant is similarly useful since it's also powered by GPT.

0

Hi Folks,

Thanks for raising this discussion!

Since AI Assistant is a pretty young product feature, it is far from ideal yet and surely, there is an area for improvement.
Please note that you may submit tickets directly to our developers via the YouTrack platform, we are happy to collect them. For anything PHP-specific, you may refer to the PHP AI subsystem, here is a current list of related reports:
https://youtrack.jetbrains.com/issues?q=Subsystem:%20%7BPHP%20AI%7D%20

If you have any specific code samples/scenarios (or even projects that can be shared privately) where you expect a better AI Assistant involvement results, it would be great if you could report these:
https://youtrack.jetbrains.com/newIssue

0

Thanks Vasiliy Yur. I was mainly trying to figure out if my experience was normal or if there was something wrong. I think my problem is that I hadn't enabled inline suggestions for all my projects. If I have any suggestions I'll post them in YouTrack.

0

this is new from AI

0

Ourmathnet 
Could you please file a new report in our issue tracker: https://youtrack.jetbrains.com/newIssue?project=LLM, and share more details about the issue you encountered? It is easier and more efficient to analyze it in the issue tracker than in a forum thread. Thank you.

0

Please sign in to leave a comment.