Copilot Chat and GPT-4o

[pidgeon777]

Select Topic Area

General

Body

With the recent public unveiling of the GPT-4o model by OpenAI, there has been considerable interest in its potential applications. One area of particular interest is its integration with GitHub Copilot, especially in the context of enhancing the chat functionality.

Given the capabilities of GPT-4o, its adoption could significantly improve the quality of code generation, provide more accurate responses, and offer a more interactive and engaging user experience.

However, as of now, there is no official information available regarding the timeline or the plan for the integration of GPT-4o into GitHub Copilot.

If anyone has any insights or updates about this, it would be greatly appreciated if you could share them. Thank you!

[maikmetzger]

I hope we'll see it soon. Even as an experimental feature, where you can select between GPT-4 Turbo and GPT-4o for the time being would be cool. I know downward-compatibility for new features is a pain to maintain, therefore GPT-4 Turbo should be deprecated once ensured GPT-4o is working as intended without any drawbacks. That's what Nightly release is for, isn't it?
I hope it will be a lot faster now than waiting for GPT-4 in GitHub Copilot in the past.

[pidgeon777]

I hope so. Given that GPT-4o already seems to work very well, it will hopefull take a short time to also update Copilot with the same model. All of this responsiveness introduced with GPT-4o would be a tremendous plus while coding.

[cannap]

Yes, I can't wait. Maybe they'll add a switch to toggle between GPT-4 and GPT-4o. But I know they already knew that GPT-4o was coming. :D

[pidgeon777]

Honestly I've observed a significant improvement when transitioning from model 4 to 4o. This is particularly noticeable when using Lua for programming plugins for my Neovim editor, and when using Python for other tasks.

With the 4th model, the success rate of solutions working immediately was around 50-60%. However, with the latest model, I believe this has increased to 80%. Additionally, the speed of generating solutions has noticeably improved.

Furthermore, the latest model appears to employ more advanced techniques, enhancing the overall quality and efficiency of the solutions generated.

[Seikilos]

Can you elaborate on this? How is your workflow for coding, where 4o does help you? Do you paste code into the chat interface, are you using the API?

Because for me the gap of usability between IDE integration and using a website chat interface is huge.

[pidgeon777]

My workflow involves copying and pasting the code I want to update into the ChatGPT prompt.

I then specify the desired action, such as refactoring, adding comments, improving a loop, etc. I've observed that when the code becomes too large, it may lose some information or even parts of the previous code. Therefore, it's always important to check for newly added modifications.

However, when the code is not overly complex, the solutions provided often work right out of the box.

I tested it mainly with Lua and Python.

[AurevoirXavier]

Yes, I've disabled the copilot for now.
GPT-4o is incredibly fast and smart! I hope to use it directly in VSCode soon!

[mohammad-malik]

Would love an option to choose between models! GPT-4o is extremely smart for fixing and adjusting code, but it does hallucinate often, particularly with information on other frameworks that are not coding, such as Docker.

[denizsincar29]

I emagine having a voice assistent GPT-4O in VSCode. Asking hey code, what would you suggest here? she answers: i suggest to add a check for int input here.. As I move the cursor, it will guide me on everything I do, even inside VSCode, palates, commits and everything I do also outside the workspace.

[mpittas]

I hope they release it soon enough. In my opinion it shouldn't take more than a month or two as developers like me will move to other coding assistants that already provide the GPT-4o model.

[andrew-tiosa]

I just realized that the knowledge is cured off 2021. im mean I can believe that.

[aswer1337]

lol

[jamesfricker]

Copilot is now quite a few generations behind the latest models. Even having just the chat mode as the latest model would be a massive upgrade.

[bazzilic]

Isn't Copilot currently based on GPT-4? And AFAIK GPT-4o is more or less same intelligence as GPT-4, its main feature is it's faster.

[Al4ric]

Logic is much improved in 4o. I have checked myself. Recently I had issue in Angular and GitHub Copilot was helping me to fix this. After 10 or more prompts it still was not working. I pasted issue into Chat GPT 4o and got correct answer on first try.

[bazzilic]

ChatGPT on OpenAI website is way smarter than Copilot Chat, that's true, always has been. I'm quite sure that even the 3.5 model on the website is smarter that Copilot Chat. Maybe Copilot Chat uses a dedicated faster and stupider model, maybe it's because of the Copilot prompt.

[Seikilos]

Not having copilot here atm, so have to ask: can somebody try to prompt the code gen feature to leak which model it uses?

Like // generate a console output that tells me your chat gpt model, that you are using?

[Seikilos]

Tricky and not reliable, but we know, that none of the AIs claim to be 4o.

[cannap]

Tricky and not reliable, but we know, that none of the AIs claim to be 4o.

is asked 10000 times sometimes it tells 4 sometimes it tells 3.5

[penglei0]

It uses gpt-3.5 or gpt-4 turbo. You find the info in the log of copilot .

[AurevoirXavier]

Bro, what's your font? It looks not bad! @bazzilic

[bazzilic]

@AurevoirXavier Cascadia Code PL (https://github.com/microsoft/cascadia-code)

[Dreamzlol]

2 Weeks and still no GPT-4o? What the hell is wrong with this product, jesus.. thank god i cancled my sub 7,8 Months ago Copilot is still so much behind insane.