JetBrains Rider Gets a ChatGPT-like AI Assistant!
ฝัง
- เผยแพร่เมื่อ 25 ก.ย. 2024
- Check out my courses: dometrain.com
Become a Patreon and get source code access: / nickchapsas
Hello everybody, I'm Nick, and in this video, I will give you a preview of the brand new AI assistant integrated into my IDE, JetBrains Rider and see how it stacks up against ChatGPT and GPT-4.
Workshops: bit.ly/nickwor...
Don't forget to comment, like and subscribe :)
Social Media:
Follow me on GitHub: bit.ly/ChapsasG...
Follow me on Twitter: bit.ly/ChapsasT...
Connect on LinkedIn: bit.ly/ChapsasL...
Keep coding merch: keepcoding.shop
#openai #gpt4 #chatgpt
On the Jetbrains blog they also state: "Access to the AI service is currently restricted to the territories where the OpenAI service is available. You can check the full list of territories here."
I was about to comment the same thing. They do mention OpenAI lol
It would be great to have an option “talk to project” when the ai can access to the whole project files. I m looking forward to see something like this
This is the only thing I want from these AI assistants. Just having them as a shortcut to ChatGPT or whatever seems like a lot of fluff for something we already have
The 32k tokens version would be needed as a minimum. But a project wide solution is almost certainly 10 years down the line , especially if it's done through cloud computing. The entire project would have to be uploaded and so version control integration would also be necessary to maintain a live copy. And for now at least, there isn't the capacity. So, this snippet based analysis is the best we can expect for some time. It would be useful to be able to set the dependencies scope, so that you could choose through a reference viewer what classes to analyse together.
@@512Squared I expect this is the kind of thing that embeddings would be used for rather than trying to fit the entire project in to the context window, but still agreed it's going to be some time before they manage to release something useful.
would be overkill though. lots of internet and computation power will be wasted. and in the end it can't even be used on large projects since they'll probs have some sort of rule in place to not send their whole source code to some AI
Fun prompts to try
- "Review the following code" to use before human review
- "Find potential problems"
- Custom abstract queries "how to..." like "Propose an ML model to play snake game" when you learn something new
- Rewriting text. "Make angry message polite and professional" for HRs or "rewrite documentation" -- either do a great job.
tip: You can actually have Jetbrains AI suggest a refactoring and apply it immediately. But to do so, you must select the code you want to refactor, then use the ctrl + \ shortcut to suggest what you want done.
ctrl + \ is specifically for having the AI generate code, which it will show as a diff. you can then ask the AI to refine what it generated or accept the suggested changes.
It would´ve been better if it could add and edit files directly. As it is now it just feels like an embedded web browser with chapt gpt 3.5 site opened.
Yeah I think that if you gonna go ahead and make your own assistant then integrate it better in the IDE
It is EAP, so lets see what features gets added before the full release.
I'm looking forward to AI coding tools that can automatically generate extracts of the code in my solution at different levels of detail so they can fit more detail into the context window and reason about the overall design and then drill down to get higher levels of detail as necessary (and ask themselves questions about their own advice to validate their information before showing it to me, CoT/ToT style, etc.)
I don't find tools like this that don't add much over me just using the ChatGPT browser interface to not be valuable. Copy/paste is very fast, and I can remove any proprietary details before submitting.
That said, they gotta start somewhere, and I'm happy that they are getting something out here to build on!
They did say that it's very early in development and the plugin will get frequent updates independent of the IDE updates.
They also said they're working on Copilot-like experience with inline suggestions as you're typing.
I‘m using it since two days for generating commit messages. I found this very useful
AI will provide great pair programming assistance. I'm not concerned about getting replaced by it.
in there legal docs, the have a list of the providers of the AI, there using, GPT-3.5 and GPT-4, from OpenAI
For what they're asking Rider is a great value proposition already, especially compared to VS Enterprise which offers little above and beyond even Community. I'd be happy to pay a little more every year for an improvement over Copilot which I pay for already.
If for some reason they include it in the main license it will be an insane value proposition
Actually, VS still has many selling points over Rider, e.g. continuous unit testing, some of debugging experience, predictable correctness and less annoying "in my face" bugs (I have met a lot of Rider bugs over the 2 years I am using it).
@@ozsvartkaroly I don't use Rider, but my visual studio is riddled with bugs. I have to close and re-open it multiple times a day to get stuff to compile. It's a mess.
@@okmarshallSounds like it's a problem with some part of your local setup. I've never had problems like that with VS even with gigantic solutions.
While Rider does indeed miss some very basic features and annoys me with bugs or breaking changes on every step.
@@LeMustacheI used to use VS Ent. but switched to Rider, some of my collegues still use VS, often hear about basic issues they are facing, though to be honest I'm leaning towards learning to use Neovim instead. Not much I'm really missing from VS, esp. not the sluggish performance, or the hanging IDE, for Rider being a Java app, it's surpricigly quick compared to VS.
So it is sending our code to Jetbrains? Unfortunately it will be prohibited in my company. I have to wait for offline models :(
In their blog post "AI Assistant in JetBrains IDEs" Jetbrains do mention OpenAI:
"At launch, the service supports OpenAI and additionally hosts a number of smaller models created by JetBrains. In the future, we plan to extend this to more providers, giving our users access to the best options and models available."
I only read the Rider specific one which didn't include any OpenAI specific mentions
It's hard to be excited after seeing Copilot X in action. Good that Jetbrains invest in this area thougth.
I have access to Copilot X, stopped using it, it was just so poor compared to the normal ChatGPT4 in practice, it always provided bad data/code/suggestions, didn't know the context properly, most is due to MS's agressive use of downgrading to faster/worse LLMs, and general worse/more agressive configuration of their GPT4 stuff (that more often than not feels more like GPT3.5 with the reponses it gives, compared to OpenAIs GPT4 API/ChatGPT with GPT4)
With rider 2023.3 you can now accept the suggested refractors and have them automatically applied. This doesn't seem to go beyond a single file though and multiple attempts to refactor the same class will give wildly different results. It feels like it's still using 3.5 under the hood which is kind of disappointing given the cost of the service.
One thing I found nice:
I recently needed a specific feature in my Jetbrains IDE which I was not sure, was fully supported in the latest versions.
I asked the AI about how to implement it and it knew exactly, what was possible and could give me an example on how to use it immediately.
That is not possible like this with ChatGPT and similar.
For a jetbrains product, you would hope for better context-wide integration. Also. I agree that GPT-4 would have a better engine for that kind of context building as a backdrop to it's reasoning engine, which I've seen in practice.
Price, it’s like 10x more if you’re a business utilising ChatGPT 3.5 v 4
Nick finally using the new UI too woohoo
I see that you use New UI, but how do you change your solution explorer to look like that? :D
I have the same question 😅
I think in the blog post they mentioned that they have their own AI provider service, which basically resends the requests to different models in different scenarios. And what I got from it, that they will have some offline models (for example for intellisense) and then online models for chat. Also in that post they had a link to page where they are saying that the AI provider service is using both ChatGPT 3.5 and 4.0
What color theme are you using for the New UI?
Thanks for this. I’ve been playing with the IntelliJ version on my channel, with similar results. It does seem pretty good at refactoring if you tell it what you want in little steps.
At the bottom of the article they link to a list of the models they are using. They are all OpenAI models. They do mention that they have their own model in the works though.
I think it could be a super valuable tool if it is able to understand context a bit better and offers an "insert suggestion" button on the response where the user has to take action if he/she wants to have the ai written code inserted into the file. It could even make use of the same code diffing view when applying suggested refactorings as you have when viewing file changes through git.
I just purchased the Jetbrains AI subscription yesterday, and Jetbrains AI assistant can do this now. Might just not have been in the product yet when it was in preview 5 months ago.
The AI assistant has a special shortcut, Ctrl + \, to have it generate code. You can simply put your cursor where you want new code, or select code you want refactored, then hit ctrl + \ and tell the AI assistant what you want done. The AI assistant will show the generated code as a diff in a new tab so you see exactly what the suggested changes are. You can then ask the AI to refine some aspects of the code it generated, click a button "accept changes" to apply it (which also closes the diff tab and brings you back to the actual file), or close the diff tab to cancel.
Sometimes giving a GPT code generator too many instructions in one prompt will dilute the result. Sometimes by breaking it into two or more prompts you get a more exact result.
For now, this is little of an improvement over ChatGPT. It would be really nice, if they made the UI similar to how code reviews currently work.
🔥
The released version of the Jetbrains AI assistant is now able to generate and refactor code using the ctrl + \ shortcut, which will show the generated code as a diff so you can see exactly what the suggested changes are. You can then ask the AI to refine its generated code, accept the changes which applies them directly to the file, or close the diff tab to cancel.
That's honestly a huge performance improvement for me. For many otherwise repetitive tasks I now just tell the AI what I want in ~20 seconds or so, have it generate the code, then do a quick scan to make sure all is in order and accept the code or refine it a little until it's just right.
Is this for EAP only?
Is copilot for business is better? It’s a good idea to compare this two
If they bring something like copilot as well, then they'll win me over.
I could image the manual copy/paste is a bit of an EAP issue. It's acutally not trivial to put a text-based response (no syntax-tree or similar) back into the correct places of your code.
Technically they didn't announce it now. It was announced some weeks back as a preview feature in (the as-of-now basically useless) Fleet editor
What's the context window for this assistant? Can I refactor a massive class?
Good tool that will be 100% restricted in most of the big tech because of NDA and potential source code leak
"It's part of EAP 6" - I thought he said "It's part of EAP... sick!". Honestly, I was thinking it too.
This service was just released. Could you do re-review to see if they made it better since this video?
As others have said, doesn't feel like a very smooth developer experience with the copy and pasting. Sure it will get better, but my issue right now is that it can do an ok job of relatively small examples like this, but even then only if your prompt is spot on.
Right now, I have to learn to write the thing to get the AI to write the code I could write myself in less time than learning the thing/trial and error prompting.
I understand that we are pretty much hardware limited for that, but i really can't be hyped until it can be fully solution-aware.... I mean it just a faster way to access ChatGPT but not a better way(Especially if it really just GPT 3.5 instead of 4). By the way, is this works offline, (So the AI magic happens on the local machine) or just simply a webserver you send the texts to?
My biggest problem with this not being solution-aware is that it right now completely useless when you (Your company) already has a big framework with a lots of big or small QOL codes for example. When you use an AI assist like this, it will generate a code that wont use the common calls your own framework has leading to inconsistent code between handwritten and AI written, which i think is a huge problem. Not to mention it can also cause problems with overridden methods and operators too as this tool would except the default behavior. (Not saying that its common or anything, just one of the many problems you can run into using a tool like this)
thoughts on your secret labs chair? was it worth the money?
This is disappointing. But hopefully they'll improve it
It is a preview after all so they can use all the feedback they can get
Is github copilot X available to everyone now or is it still in waitlist only?
@@sergeiburtsev5712 This is not the X version. That one has a separate beta with a waiting list. Source: I use the $10/mon beta and am in the X beta. They are not the same product - very different in fact.
It would be be nice if I could use my own chat gpt subscription
A quick win will be to generate xml documentation without leaving the IDE.
does that mean that GitHub copilot x won't be coming to rider officially ?
I’m sure it will but GitHub will make the plug-in, not JetBrains
Not sure where to put video suggestions but here’s one. Speed comparison of http vs grpc and how to best write grpc endpoints.
Great idea! Thanks
Does it use ChatGPT 4 or 3.5?
Microsoft hear your voice about killing the visual studio for mac, and they annonsing end of this IDE.
LTT probably called it. We'll have a move back to mainframes and self-hosted AI. Or we'll continue this dystopia where privacy violation fines and cloudbleed are calculated as a cost of doing business.....
Where can I get new Rider?
I downloaded it from the JetBrains Toolbox
@@nickchapsas Thanks, I just updated and switched to new UI, looks cool :)
Has the AI Assistant improved since, @Nick?
i works great for helping writing unit-tests
Open source, commercially viable llms exist. OpenAI was an og player but they’re far from the only ones now
It would be interesting to compare it to a similar free alternative Codeium, which has both inline suggestions and the chat, and it has a plugin for jetbrains IDEs.
When it’s officially released I will definitely have an AI face off
No idea about rider on WSL but intellij is a *complete* trainwreck. 32g and it demands more when vscode is using < 1g on the same project.
Rider is more an alternative to VS, than VS Code, so hardly a fair comparison.
@@michaelnjensen I'm asking if rider is essentially broken on wsl too.
what's that UI are you using on mac or something ?
It’s just Rider’s "New UI”
@@nickchapsas looks really nice thx I ll try it
search for "Running Large Language Models locally - Your own ChatGPT-like AI in C#" by Kent Sharkey
I'm not going to be impressed before the entire solution is "in context" of what it does at all times. Because unless I can discuss the code like I would with a colleague,no benefit for me
It would be also important to mention that developers should be aware of where they should use this kind of tools and where not. Meaning everything that is being used as input is actually going to Rider’s Server and it should not contain any sensitive information or intellectual property.
So basically we are all gonna be AI trainers, training the AI how to code by uploading our code "anonymously". Further powering this thing, effectively replacing our own jobs lol. I guess devs don't like having nice salaries lol
I hate AI in my IDE. It just makes it slower with bad suggestions.
Fir name is a typo... ? What a dumb AI. I obviously only allow users in my app those have a named fir on their backyard.
As usual this seems to not actually understand things, can produce errors, and practically requires you to have the knowledge you would have needed to write or refactor code in the first place, but now you get to waste some GPU hours on a rack somewhere while doing it
using codewhisperer because its free
ελληναρας
They keep copying Visual Studio
Copying of course, but also improving
Switched from VS year ago, and never looked back
I'm happy about that, rarely open my VS Ent these days, Rider is just better on all areas, esp. speed and stability compared to VS, esp. if you have many diffrent solutions open at the same time.
@@michaelnjensen I think you are not being sincere. lol
First!