I currently have a list of almost 20 ideas for personal (code) projects I would like to develop some day. Most of them are plugins for or integrations between productivity tools I currently use, even though two or three of them are more complicated - for example, workflows including categorizing natural language texts or processing images.
In the last months/year I haven't prioritized starting any project from this list because I'm trying to limit indoor hobbies in my free time, and because I'm unsure whether any of these projects would pay off in productivity the time I took building them. Also, in my experience in previous projects, I sometimes become too perfectionist/overengineer things, and end up loosing interest in the project before I get anything usable.
Lately, I have been thinking of reevaluting this logic in face of new tools like Auto-GPT, gpt-engineer and SuperCoder, which promise to greatly reduce the costs of kicking off a new application by using LLM-based AI "agents" to do most of the coding heavy lifting.
This market resolves as YES if, by the end of the 1st quarter of 2024, I open source a new repository in my personal GitHub that both:
was authored with major contributions of a LLM-based, semi-automated coding tool like the ones mentioned above - even if I have to do some final polishing (I will point that out in a
Credits
section of theREADME.md
file)have at least one tagged release in the repo, signaling that I think the result project is minimally reproducible and useful for other people.
I won't bet on this market.
🏅 Top traders
# | Name | Total profit |
---|---|---|
1 | Ṁ60 | |
2 | Ṁ41 | |
3 | Ṁ22 | |
4 | Ṁ18 | |
5 | Ṁ6 |
@vluzko the precise line is somewhat subjective, but I would say a "major contribution" from an AI agent should include:
Most of the boilerplate code (authentication/authorization workflows, set up of servers & database connections etc)
Implementing some of the business rules, even if it is not at 100% accuracy.
Say, for example, one of the simplest ideas on my list: contributing a community n8n node that reads from ActivityWatch (AW) data. I expect an AI not only to set up the basics of a npm package and development environment (dependencies listing, linting, gitignore etc), but also to be able to interpret the documentation for both APIs and draft the operations and fields needed to list AW buckets and read existing events.