How would you use Github Copilot?

To create a challenger bank that only uses float for their money calculations 😍

I haven't tried it yet, but I think the most you can get from a tool like this is TEMPLATING.

You will still have to know ALL the engineering behind creating an app. It will take decades before a business man with an idea could build & deploy with just AI. It's not even an alternative to no code tools.

I dont have access to it but one of the devs at my day job has access to it and said that it is pretty good.

Since I may not have access to it for a while I decided to give https://tabnine.com a shot and have been using that for a week.

The verdict is that it is very underwhelming.

Its a shame and I wanted to like it given how little chance I stand to have access to Copilot but it's the unfortunate reality.

Tabnine is not a bad product, I would say that it gives me a good suggestion 10% of the time but it does not really justify the $15USD/month price tag.

It's a bit like comparing a Renault Zoe to a Tesla. Once you have seen what a Tesla can do and there is no going back to something that does not perform as well.

I have tested Tabnine with typed codebases and non typed codebases and the results are pretty similar.

I would have assumed that the typed codebases would give Tabnine an edge but it barely makes a difference.

I will continue using the extension for the next 3 weeks and most likely will cancel my subscription as the benefits are marginal at best.

In essence what I foresee happening here is a winner takes all approach, why would you want as a dev use anything over Copilot if Copilot is at the same price point or even more expensive than Tabnine/other AI tools in this space?

To catch up to Copilot, you would need first access to GPT-3(Tabnine uses GPT-2 with some custom models) which hard to get this days, two you'll need access to as much code as possible, then you need the cash that will pay for all the engineers that will work out the bugs and rough edges over time.

That's not even getting into the legal side of this thing where you will have to defend your product against some (potentially legitimate) uproar from the dev community and people suing you for copyright infringement.

In reality if Copilot becomes a public product that you can subscribe to, then Tabnine will be obliterated.

It is just not possible to compete against it.

Only another GAFA could enter this space and give Copilot a run for its money but the problem is that MS controls Github and I doubt that MS would simply let Google, Apple or Oracle have access to all the open source repositories knowing that this access would be required to create a potential competitor to a business that could easily be valued at billions of $ within the next few years.

Tried it for a bit, amazed by what I can predict, UX of how it predicts needs to be improved.

Having to check its prediction really takes me out of the flow of programming. I’m slowing down because I’m reading prediction.

If I don’t read the predictions, copilot will now and then make a tiny mistake (for example assuming I want to use .width vs .x of a rect object), those mistakes aren’t obvious if you miss them, thus taking more time to fix.