Resonance of Tokens: Selected Iterations
Автор: Shin Seung Back
Загружено: 2026-02-18
Просмотров: 16
Описание:
How does an AI, trained exclusively on text and images, imagine sound?
This compilation features 30 tracks generated by six different large language models. Each model translated written descriptions of five elements (Earth, Water, Fire, Air, Human) into audio via Python.
No music software, no samples. Just raw frequencies mathematically calculated from scratch. The cover art is also purely generated SVG code.
[Tracklist]
Anthropic: Claude Opus 4.6
0:00:00 Earth
0:00:08 Water
0:01:08 Fire
0:01:38 Air
0:02:03 Human
DeepSeek: DeepSeek V3.2
0:02:36 Earth
0:02:40 Water
0:03:10 Fire
0:03:25 Air
0:03:35 Human
OpenAI: gpt-oss-120b
0:03:47 Earth
0:03:55 Water
0:04:25 Fire
0:04:37 Air
0:04:49 Human
MoonshotAI: Kimi K2.5
0:05:19 Earth
0:05:27 Water
0:05:57 Fire
0:06:07 Air
0:06:15 Human
Z.ai: GLM 4.7
0:06:27 Earth
0:06:35 Water
0:06:50 Fire
0:07:00 Air
0:07:10 Human
Google: Gemma 3 27B
0:07:24 Earth
0:07:29 Water
0:07:39 Fire
0:07:44 Air
0:07:49 Human
Notes:
Claude Opus 4.6: With adaptive thinking, this model required more generation time than the other five models combined.
Exclusions: Models trained on audio data (such as GPT-5.2, Gemini 3, and Grok-4) were excluded.
Explore the full project:
https://shinseungback.com/lab/project...
Повторяем попытку...
Доступные форматы для скачивания:
Скачать видео
-
Информация по загрузке:
