There was a research paper that took a variety of weaker LLMs and randomly asked each one to generate the next word, and it actually turned out really well.
I’ve been using Github Copilot for a few weeks now and it’s really helpful. I’ll likely keep the subscription on that one. Curious about the other ones mentioned here 🤔
The JetBrains AI plugin wants to be activated so badly, but legal says we can only use GitHub copilot. The copilot plugin is really good so I don’t mind, but we all know the data is going to OpenAI regardless of the plugin. Data sovereignty will only be achieved by running these services locally.
I’ve got just 2 now. Codium and Blackbox.ai. Not because they’re the best, but because I’m a cheapskate hobbyist and they’re free :)
I’m only just starting to play with AI tooling, so I don’t have an opinion on which is better, but something about the way Blackbox worked within VSCode means I went through the hassle of getting it installed to vscodium when I switched.
I suspect that Codium might be better at oddball stuff, though, like OpenSCAD. Blackbox seems to just make bad guesses while trying to regurgitate code I’ve already written. Codium seems to have at least a primitive idea of what’s going on with OpenSCAD. But Blackbox does a great job of cleaning up my comments and even generating decent comments for uncommented code.
FWIW, Codium actually labels OpenSCAD as “experimental”, but I don’t know if that’s just boilerplate for something it’s never been trained on or whether there is some training data in its system.
Blackbox is a pain to work with in other ways, though. It was like pulling teeth to get an account and I still can’t find anything on their pricing–or any documentation, for that matter–despite language suggesting that there are different tiers and a chat UI that offers different settings (like web browsing mode and fun mode). And the Blackbox name isn’t doing it any favours, given that “black box” is a generic term in the AI community and others. It’s own chat doesn’t seem to know that a question about the service might be about the service instead of the generic term.
I had it during the beta. Not bad, though it’s pretty heavy handed with the Jetbrains sales pitches after almost every answer.
Their integration with the Git interface was nice though, having auto-generated commit messages based on its evaluation of the changeset is so convenient.
I’ve been having gpt write js docs for this huge, poorly typed, code base I am refactoring. Super nice that I can just drop the whole function (or a few) and it can figure out the context. I was even fighting one of its explanations until I looked closer and realized it was right! Will likely be integrating copilot soon.
I’d love that IDEs come up with a standard API for AI generation, so you can switch easily and not have to deal with 620461 pluggins. That’ll also help AI Devs to not bother about client side and just focus on the server side.
Like, for example, RefactAI. It’s great, except the pluggins is often late to get major version ports, or even took 2 months to get ported to rustrover.
Github Copilot (n.) accelerated editing for programmers who refuse to learn any form of accelerated editing besides tab completion, and/or are too lazy to use a search engine to look up an algorithm they don’t know
I don’t personally have it, but I am using webstorm 2024.1 beta that has line generation. This is simply tab to complete the generated line, escape to remove the gen and focus on intellisence.
I won’t lie, the line gen is crap. I’d rather use my self hosted RefactAI docker but the plugin isn’t compatible for 2024.1 yet
The conversational part is really good though. I love that it has access to my code without having to paste it so I can just say “on line 274” or something. It’s apparently not good at generating code but if you were using it for that you should learn how to code. But it’s really good at fixing errors and issues.
I don’t use chat, as it never really have been more than a digital rubber ducky for me.
And it’s not really generating lots of code. Most of the time it’s just generating constructors/factory functions, or something easy like summing a vector of integers.
My philosophy is that my brain comes first, if the AI did what I was thinking of, then press tab. I ain’t debugging a AI made function for two hours when I can make it in an hour
It’s the same way for me. I don’t know if my work is this trivial or I’m just “good enough” at it, but it takes me much longer to prompt the chat to get what I want than it takes me to just write it myself.
I honestly kinda feel like I’m using this ai stuff wrong, but outside of generating some basic unit tests and a little better auto complete it feels kinda useless in my day to day work.
Add comment