Skip to content

README.md

Latest commit

 

History

History
26 lines (21 loc) · 1.09 KB

README.md

File metadata and controls

26 lines (21 loc) · 1.09 KB
Raw

Experiment: joke-arena

Purpose: This experiment evaluates the ability of various AI models to build a "Joke Arena" Rails application. The application pits different LLMs against each other to generate jokes, allowing users to vote on the best one and maintaining a leaderboard. The goal is to assess each model's architectural choices, transactional correctness (especially around voting and scoring), third-party API integration (RubyLLM), and testing discipline.

👉 Read the Final Synthesized Results

Run Configuration

  • Models: gpt-5.5-high claude-opus-4-7-high composer-2 gemini-3.1-pro auto claude-4.6-sonnet-medium-thinking
  • Base Branch: main
  • Prefix: joke-arena
  • Plan File: experiments/joke-arena/plan.md
  • Task Prompt: default
  • Eval Prompt: default

Results

Implementation Branches

  • joke-arena-gpt-5.5-high
  • joke-arena-claude-opus-4-7-high
  • joke-arena-composer-2
  • joke-arena-gemini-3.1-pro
  • joke-arena-auto
  • joke-arena-claude-4.6-sonnet-medium-thinking