Share your opinion about Codex vs ChatGPT

API
1

I found ChatGPT to be more helpful in assisting me building my projects than Codex.
Codex made more mistakes than ChatGPT.
Did something similar happened to you as well?

3 Likes
2

Yes.

I find that if you want to create a draft method, ChatGPT creates “pretty good” first draft code blocks which are sometimes almost perfect and sometimes “way off” but if you keep the code blocks small, which is a good coding practice anyway, ChatGPT is very helpful more than other models.

For the codex, I use it as a the core model in the Visual Studio Code Copilot extension and the same can be said for codex code completions, sometimes right on target, other time annoying.

I find both very useful, TBH. It depends on where you are in your coding workflow.

For example, just now having a coffee, I was writing code to fine-tune fine-tuned models which pulled all my FTMs from the API and put the name of the models in an array.

The code completions in VSC were very helpful. For example, I created this module method:

 def self.get_array_of_fine_tuned_models
        client = get_client
        response = client.finetunes.list
        output = []
        response["data"].each do |ftm|
            output << ftm["fine_tuned_model"] if !ftm["fine_tuned_model"].nil?
        end
        output
    end

… a large part of the code above was helped along by the OpenAI codex code completions, saving a lot of keystrokes. I like saving keystrokes, as we all do!

Then I took that code and wrote some embedded Ruby to create this form, but since it was a modified copy-and-paste from a similar form, no code completions were needed.

Screenshot 2023-02-14 at 10.07.22 AM
Screenshot 2023-02-14 at 10.07.22 AM2088Ă—1140 173 KB

However, when I created modules for various methods to correlate embeddings, I used ChatGPT to draft methods (which of course need to be checked and not taken for granted), like so:

Screenshot 2023-02-14 at 10.10.12 AM
Screenshot 2023-02-14 at 10.10.12 AM1688Ă—1126 139 KB

Blah,blah…

So, I like the codex for code completions in VSC and ChatGPT for drafting new methods.

After all, ChatGPT is basically a text-competition engine, and they both have their places in a developers workflow and both increase coding productivity significantly.

Hope this helps.

6 Likes
3

Agreed, I write code for PLC’s and at the current stage ChatGPT writes better code than CODEX.

2 Likes