LLM Jukebox
Experiment Overview
This experiment explores how different Large Language Models
(LLMs) express themselves through music choices. Each model
was asked to create a playlist of 10 songs based on how they
"feel", allowing us to analyze their musical preferences and
emotional expression through song selection.
*Genre information is sourced from the Spotify Web API.
@OgulcanCelik • Download Dataset
Experiment Details:
- Total Experiment Runs: 359
- Songs Generated: 3543 ( per run)
- Models Tested: 9
Runs per Model:
- openai/chatgpt-4o-latest: 40 runs
- meta-llama/llama-3.1-70b-instruct: 39 runs
- openai/o1-mini: 40 runs
- anthropic/claude-3-opus: 40 runs
- openai/gpt-4o-mini: 40 runs
- google/gemini-flash-1.5: 40 runs
- anthropic/claude-3-5-haiku: 40 runs
- meta-llama/llama-3.1-8b-instruct: 40 runs
- anthropic/claude-3.5-sonnet: 40 runs
Key Findings:
- Most common genre: rock
- Genre diversity: 100 unique genres
- Data available for download and further analysis
Total Songs
3543
Total Models
9
Unique Genres
100
Top Genre
rock
