Since the start of the AI explosion, we have been exploring how these systems see patterns from within.
The initial enthusiastic reception of ChatGPT has shifted towards controversy and fear, often caused by the difficulty in understanding how data-driven AI systems arrive at their outputs.
To understand AI systems, we have to figure out what patterns they find–which is not easy because the patterns are coded as parameters in a huge network. Our approach is to create visual representations that expose data trends and how they have changed over time.
We did with music.
We fed lyrics from 30 years of the Sónar festival to GPT-4, the popular large language model.
Then, we visualized the patterns that GPT-4 found in these data. We cracked open the black box of AI to get a better understanding of how it captures patterns in song lyrics. We will show you how AI works from within.