OpenAI Codex FAQs

[Harish]

Two questions

• The Tool that is used for preparing the demos for Codex - Codegen, would that be available for the community to use or take a look at?
• Is there a separate content-filter for Codex API or should we use the same one as for GPT-3?

[protein_patty]

I don’t have access yet. Could I just ask? What if someone were to type “Make a Pac-Man program.” Could Codex do that? You know, a playable version? Just whip it out? Just wondering what this thing can do. (I saw on the Internet that someone had typed something like “Make a website for Twitter that’s like Apple.com” and Presto! It just magically whipped it out and apparently it looked awesome. Not sure if it was this AI or a different one.) Thank you.

[Harish]

I think it pretty much did it(here is more code beyond this screenshot)

t

[protein_patty]

Can you run it? Does it work? Can you share any screenshots of the game screen?

[Harish]

I got access few hours back.

Yes, they show up in the engine list in the playground.

Available via API also

[u2084511_felix]

@Adam-OpenAI

What programming languages does OpenAI Codex know?

OpenAI Codex is most capable in Python, but it is also proficient in over a dozen languages including JavaScript, Go, Perl, PHP, Ruby, Swift and TypeScript, and even Shell. It also has knowledge of popular APIs and libraries.

In yesterdays live demonstration it was revealed that codex currently has a 37% accuracy when generating code. Assuming this is codex at it’s most capable (Python), how accurate is it when coding in other languages?
I’m particularly interested in C++ as this is the language I use for games programming.

@Adam-OpenAI

Can I fine-tune the OpenAI Codex model?

Currently, the Codex model cannot be fine-tuned via the fine-tuning API.

Can I train the OpenAI Codex models to understand additional languages?

At this time Codex users cannot fine-tune the OpenAI Codex models for other languages. If there is a language you would like the model to better understand, please contact support@openai.com with details about the language and your use case.

I have already emailed an appeal for greater C++ training, but I feel this was probably more for those already using codex, as the responses felt a little automated and assumed I was just begging for access (which was only half true lol). Obviously, I don’t yet know how good codex actually is at C++ because I haven’t used it, but not seeing C++ even mentioned in the list of languages was what prompted me to send an email.

I’m sure others would be interested to know more details about language proficiency too, so I thought it would be better to carry on my codex for C++ campaigning here, hehe.

[Harish]

Not yet. I have used only through Playground and short snippets through Jupyter notebook so far.

[Datasculptor]

congratulations! have fun : )

[berkayalanpro]

Thanks for all. Will it be any tutorial for Codex?

[berkayalanpro]

I would like to learn how to use it and how to build the application architecture, its components, going from idea to running application.

[dara]

Thank you for considering us for the Codex. I have an issue which I cannot find an answer in your organization’s documents.

We need to build hybrid net solutions, while you provider a powerful tool for certain applications, it is not expansive to other computing niches which we need to deliver services for.

What is the mechanism for us to mesh up your e.g. Codex neural nets with native ones we have developed? Often we have non-Neural Net learning algorithms which we need to use for certain applications.

While the technology is quite advanced we cannot use it as a monolithic platform.

That being said, could we talk to anyone at Open AI with enough technical acumen to discuss in person?

Dara

[mateo.lack]

I can’t actually remember if I submitted the form for the waitlist, does it email you when you submit it? Maybe I should just submit it again?

[zarugeos]

If you really can’t remember then submit but if you do there is no reason to submit again

[KM]

Does anyone know the difference between davinci-codex and davinci-codex-v2?

[ricwo]

@Adam-OpenAI Quick question regarding token limits: Is the rate limit of 150k davinci tokens / minute applied to the number of generated tokens, or to the max_tokens parameter that’s passed to the API call, regardless of how many are actually generated?

I hit this limit very quickly with a small number of one-line completions (I do, however, pass 2048 as max_tokens).

[ricwo]

It’s in the FAQs in this thread. Also I frequently get an error response from the API which says

openai.error.RateLimitError: Rate limit reached for tokens. Limit: 150000.000000 / min. Current: 162320.000000 / min. Contact support@openai.com if you continue to have issues.

[bellocchiocarmela]

An error occurred, please report this issue if it persists. Does it need credit card or is optional?? I can’t access to sandbox!! Help me please!!

[James101217]

Where ia the openai?

[James101217]

Where is the URL for this image? I can already start using codex-javascript-codex, but I don’t know where the url is for this image. Hope to reply to me. pls

[loa]

How many parameters are in davinci-codex and Cushman respectively?