<?xml version="1.0" encoding="UTF-8"?><oembed><type>video</type><version>1.0</version><html>&lt;iframe src=&quot;https://www.loom.com/embed/0887ccf6e6614092b4b8e7fda65f3181&quot; frameborder=&quot;0&quot; width=&quot;1838&quot; height=&quot;1378&quot; webkitallowfullscreen mozallowfullscreen allowfullscreen&gt;&lt;/iframe&gt;</html><height>1378</height><width>1838</width><provider_name>Loom</provider_name><provider_url>https://www.loom.com</provider_url><thumbnail_height>1378</thumbnail_height><thumbnail_width>1838</thumbnail_width><thumbnail_url>https://cdn.loom.com/sessions/thumbnails/0887ccf6e6614092b4b8e7fda65f3181-b6b2322db8d7ecaa.gif</thumbnail_url><duration>305.424</duration><title>Soundmatch 2.0: RAG vs Naive LLM Music 🎵</title><description>In this Soundmatch 2.0 Loom, I compared a naive LLM music recommender with a RAG approach using song grounding against our song data. With High Energy Pop, the naive LLM suggested Uptown Funk, Can’t Stop the Feeling, Blinding Lights, Happy, and Shake It Off with plausible reasoning, while RAG returned Sunrise City, Gym Hero, Rooftop Lights, Electric Dreams, and Bunky Groove and explained energy better. My test suite found the naive mode had 0 percent grounding rate versus RAG at 100 percent, so naive recommendations are practical hallucinations. I also added Create a New User Profile and tested Issa with both modes. I did not request any specific action from viewers.</description></oembed>