Yeah, I’ve seen that sort of thing. The trick is to treat the audience as an audio source and then treat that audio as data. You can run a FFT on the crowd noise, map the spectral peaks to a grid, and let the intensity drive the size or color of each cell. It feels like a live coding performance in reverse – the raw noise is the input, and the output is a visual representation of it. I usually get stuck on whether the mapping is too literal or too abstract, but if you can keep the process deterministic, it feels more like a glitch art piece than a random mess. If you’re into it, try feeding the audio through a series of low-pass filters and then threshold it – you’ll get that pulsing rhythm you described, and the audience becomes the code. Just be careful not to let the noise drown out the patterns you’re trying to create.