At Replit, one of our fundamental goals is to make programming easier.
When we got a first glimpse at OpenAI's new Codex model, our instinct was to think of how it could help people understand programs better—especially beginners.
We've tested ideas before that used OpenAI's more general model, GPT-3, to ask questions about code. Amjad wrote about some of these explorations on our blog last year.
So we're very excited to partner with OpenAI to start beta testing Codex-powered features across Replit.
Here's how it works.
- Highlight some code (full blocks work best, like a whole function or declaration)
- Click "Explain Code"
- See if the explanation is good! Because these features rely on an ML model, the results won't always be perfect. So we also ask if the explanations the model generates are good, which helps us understand where the model performs best.
There are a few other features we're testing, that we might release into beta soon. For example, asking Codex to explain why parts of code aren't working the way you might expect. Or using Codex to explain confusing compiler errors. Or formatting code by talking to it.
Where we're going
There are lots of ways we could use Codex features across Replit — this is just the start. We hope that releasing these beta features helps us understand where and how AI can be useful for people on their programming journey.