There's been some major news in the AI coding space this week:
Stability AI (the company behind Stable Diffusion) has launched StableCode, an open-source model to help users generate programming language code.
StableCode will support development in Python, Go, Java, JavaScript, C, markdown and C++.
In other news, Google has announced the launch of Project IDX, a new AI-enabled browser-based development environment.
The IDE will be based on VS Code with all the AI-code-generation quirks by Google.
Will this make indie hackers (like us) faster in things like developing minimum viable products? So far, code generation has been hit-and-miss. For example, this recent study showed that ChatGPT's odds of getting code questions correct are worse than a coin flip.
Yeah I am yet to see a convincing argument for how a large language model can understand the full cross-file context understanding required for full-stack application development. By its nature it just starts getting more and more confused as it deals in averages and density of correlation whereas code needs to be explicit.
Having said that though - it's definitely progressing fast. Maybe these next rounds of tools will close that full-stack app gap. It's amazing for simple refactors, snippets and improvements on a page or segment of code so it'll be god tier when it gets to that next point.
My experience with AI coding assist is that the AI doesn't seem to take into account the actual functions and structures in the code. I routinely get code suggestions that are clearly based on open-source code you can find on GitHub rather than the specific code that I'm writing. This is true even if the function is present in the file I'm working on.
I've built an expenses tracker bot on telegram using chatgpt. I never would have imagined doing it myself in a week. Sure it gave bugs initially, had to do some manual work but it was helpful me in increasing my productivity I must say. Now things are bit better with gpt 4's plugin and interpreter
I gotta say as a developer I don't believe AI can take over developers. Sometimes understanding what clients want is harder than the actual coding. AI right now can't get everything people want. But... can help a lot during developing.
I develop mostly in Python, using Visual Studio Code. I have LOVED using the new tools that have come out from Github: copilot just blew my mind. This tool has allowed me to develop a few chrome browser extensions, an area I had not developed in before.
ChatGPT can be hit or miss for software development.
Also, really awesome tool I love: Phind
AI's capabilities continue to expand! With the ability to generate an extensive amount of code, it's opening doors to efficiency and creativity in ways we've never seen before.
In my experience, AI coding tools are amazing for generating boilerplate code that requires very little thought. However, they struggle with more complex sections that involve intricate logic. I absolutely love this aspect, as it gives me complete control over the critical components, while I'm happy to delegate the mundane and simple tasks to the AI tools.
AI tools like copilot or Phind also help a lot when navigating an unfamiliar code or working with an unknown library. They provide very reasonable hints. It's often easier to double check the correctness of the code than to write it from scratch (except the cases where the generated code is 99% correct.. in which case it's very hard to debug).
Currently working with Tabnine AI (linked with my IDE).
It does decent work in the scope of single files. Looking forward to seeing those tools go beyond and build an understanding of the whole module/project, or even 3rd party deps I utilize
Great Tool
Yes, AI can help in generating more maintainable and clear codes for development. However, to get maximum benefits we can balance the involvement of manual efforts and Chat GPT.