[2024-10-01]
Lately, I’ve been interested in the different survey/voting methods, the most curious of them to me seem Elo ones. Thought it’d be a good opportunity to check out how ~1yr of AI progress improved the capability of software generation from one go to have a demo to play with.
Long story short, I wanted to see if I’d be able to have (not create) some web service with realisation of basic Elo algorithm (wiki link) in quick and easy way and it worked.
And Claude generating the baseline and Replit agent finishing the job did do the trick - with few attempts and roughly 1.5hr of final effort there’s a deployed service that’s able to generate a leaderboard from a relatively big set of choices, 1k options in my case - https://landmark-comparison.plvch.com/
Tech Process:
The whole process takes the short prompt in Claude:
The caveat, as with earlier versions of ChatGPT it probably won’t be able to have a state (i.e. updated scores), some database connected, etc. It’s good at generating some static examples or code that executes somewhere else.
But as the primary goal is to create a playable prototype, I’ve taken the baseline code, put it into another platform (Replit - interactive IDE on strong steroids) and asked the agent to add the dependencies needed to get a fully operating version.
And it managed to do it, rewriting initial Js in Python, but with the working connection to initial data and progress state.
Overall, on the experience: