Hacker News Reader

Comment by tyoma | Hacker News Reader

In my testing, Codex is still better at some code tasks than GPT 3.5/4. I am sad to see it go.

It looks like Codex will still be available via Azure: https://learn.microsoft.com/en-us/azure/cognitive-services/o...

I imagine MSFT would never give their customers 3 days notice of imminent service shutdown.

The problem that I saw with Codex is that it routinely bugs out and starts repeating something infinitely. And the model parameters can't stop that from happening.

I see something similar with ChatGPT too... it seems to forget a previous constraints and jumps to the generic. In my example, I was trying to generate an ffmpeg command line that given a source file will, during transcode, generate stereo audio even if the source file had no audio (so basically output a blank audio stream). It kept using 0:a in the complex filter and I would ask it what would happen if there was no audio stream in the input file? It would actually correct itself but something else would break in the fix. I would point out something else was wrong and it would fix that but reintroduce 0:a.

Maybe the output was long enough that the context window would miss your previous prompts?

I'm late, but the model can remember 4000 tokens, which according to openai is about 3000 words

This feels like an aggressive push to force codex users to adapt to GPT3.5/4. I'd imagine they will be integrating some of codex's features into GPT4 soon.

It's ridiculous. Now we all have to implement "chat" even if we all want is code completion

No. I have an app that uses both. You do not have to implement chat.

All you have to do is 1) call the chat completion API, 2) structure your prompt in JSON as system+agent+user, and 3) get the completion from a different JSON node. You don’t have to round trip or maintain context or anything.

I'm seeing that the chat api wants to always wrap its response in "```python ```" when it generates code. Is there a prompt trick to stop this, or should I just be stripping it manually

Azure is good for its years-in-advance warnings of shutdowns

Reply on news.ycombinator.com