Phlo Young

Phlo Young teaches practical workflows for AI-assisted music creation and dataset exploration. His published materials explain how to turn prompts and existing sounds into music, and how to inspect conversation datasets before fine-tuning a model. He presented “AI Music Generation: From Prompt to Production” at AI Engineer World’s Fair 2024 and publishes on GitHub as YoungPhlo, also using the name Phlo and X handle @YoungPhlo_.[2][3][4][7]

Young’s music workshop materials distinguish three approaches: style transfer and voice conversion, text-to-music, and audio-to-music. The examples show how existing sounds can supply creative starting material: beatboxing becomes drums through Stable Audio 2.0, a dog’s sounds become a lo-fi remix through Suno, and finger drumming becomes punk-rock guitar through Udio. Young assembles these examples to explain the different workflows rather than presenting the underlying models as his inventions.[4]

Young questions the label “AI Music” when a human uses AI as a creative tool, suggesting “AI Assisted Music” as a more accurate description. His workshop notes acknowledge the fun of meme-song generation while arguing that professional production requires more control: “Producers want more control. Will create at a greater pace given more tools.” He predicts that generative tools will become part of production much like digital audio workstations and plugins, while maintaining that “The human connection in music triumphs.”[4]

His prompting template makes musical instructions explicit. It separates a song’s theme and mood from style tags, then organizes lyrics and musical or vocal directions into an introduction, verses, choruses, an instrumental break, and an outro. This gives participants a structured way to specify a song rather than relying on a single broad genre request. In his presentation slides, Young also lists a personal use case: creating songs to help his daughter memorize facts for school.[4][5]

On May 6, 2025, Young appeared on Tool Use — AI Conversations in “How To Turn Text Into Amazing Music (ft Phlo).” The episode covered Suno, Udio, and Riffusion, creative control, and a live Suno song-generation demonstration. Latent.Space’s June 2025 World’s Fair recap also credited him among the event’s interviewers.[6][1]

Before his 2024 music workshop, Young published materials for an AI Engineer Fall 2023 talk on visualizing datasets. That earlier work provides a separate example of his practical AI teaching: examining training data before using it to fine-tune a model.[7][2]

Young’s dataset visualization tutorial adapts a workflow that credits BERTopic and an original notebook tutorial, including Maarten’s Llama 2 tutorial. It combines the turns of each conversation into one string, prepares SentenceTransformer embeddings, uses quantized LLaMA-2 to generate topic labels, and displays topics in tables and graphs. The stated aim is to reveal themes and unexpected conversations while keeping datasets local, without requiring uploads to third-party services.[7]

Young explains his implementation choices and their limits. He prefers BAAI/bge-base-en-v1.5 for its balance of quality and speed on his machine, and advises experimenting with BERTopic’s min_cluster_size because the appropriate setting depends on the dataset. His preparation steps include checking conversation lengths and truncating inputs to fit the model’s context window. He also warns that extremely large datasets may not fit and that nondeterministic models can produce somewhat different results across runs.[7]