Anyone used the AI service yet? Can you tell us what it's like and give your opinion?


I am never going to be able to use the AI service because there's not an option for it to be locally hosted AFAIK although confidential side agreements may exist between Jetbrains and individual companies (I say this because some largish part of Jetbrains' clientele, like the financial services industry,  are in the same boat as I am). 


I was wondering two things. One is, has anyone used it and if so can you give your impressions of it ? What does it do ?  What does it do well ? What annoys you? How do you use it and how has it helped you?  


The second is does anyone have awareness or an opinion on self hosted alternatives  specifically, or otherwise,  for coding? 



Thank you !!



Hello, java developer. Sorry to hear you can't enjoy our AI Assistant yet. 
However, as far as I know, we have plans to support local models in the future; the supported feature set will most likely be limited. Please consider upvoting this feature request to show your interest and receive updates and leaving a comment if you have any additional thoughts about this matter.


Thanks I did upvote that. 


I have bought license and I need to wait for almost 8000 years.


Paweł, it looks like you met the problem described in this issue in our YouTrack, specifically the second case described in the Cause paragraph. Please submit a request to us using this form, and we will help you with the problem. 
Also, please note that the possible cause could be the following: 


Sometimes though, we see that users are chatting in the same chat for every new question, which results in a context growing and could cause abnormal usage detection mechanisms to be triggered. We are analyzing how this UX aspect can be improved.


If this is the case, please consider changing your way using the AI Assistant chat.


OK I tried it on my non-dev machine with a toy problem. 


I asked it to solve a very moderately difficult problem and it said a lot, repeated my question back to me, injected some true but irrelevant info and came up with the lamest correct answer there was. OTOH the other online AI I tried gave me similar lame answers but seemed much more “lifelike” , didn't inject irrelevancies and gave me 2 more variations on the same type of lame answer. 


I still think it's probably worth it though because it can catch stupid or subtle errors, make good suggestions (I didn't try that) , tell you things you legitimately didn't know and probably explain ad hoc question on language features you keep forgetting the details of (or  you never learned given the new galloping release schedule of Java). So it's a good basic work mule but, hey,  that's work I didn't have to do, just like a mule carries things so I don't have to.  I'll bet it rocks boiler plate like nobody's business.  


I didn't know how sophisticated  a question it might be able to reason about but now I am sure,  my job is safe. So it seems OK to me for that reason too. 


You have to keep in mind how these things work. LLM and transformers do one thing- they predict the next word to write. They have zero understanding of what they're doing or why and have absolutely no mental model of programming or anything else. If the question you're asking is not a lot like (really, nearly identical to) something which's been asked on line (the source of the text they train these things on) you're going to get marginal results. It's not creative like that and doesn't “think” or reason.  The refactoring code in IntelliJ has a more sophisticated model of what a  programming language is than an any LLM does because the LLM has exactly no model at all. 


A couple of famous quotes come to mind. 

First, Feymann's observation that quantity (of the CPU cycles and the corpus of language it processes) has its own quality. 

The second is what Dr. Johnson said about a dog which has been taught to walk on its hind quarters- "It's not that the thing is done well, but you are surprised to find it done at all."


Please sign in to leave a comment.