Just for curiosity, I've saved ~30 seconds from one of the videos and asked ChatGPT. My usual go to is Claude but it looks like free version does not support analyzing audio files. Anyhow, ChatGPT says:
There are 230 detected note attacks or transients. The standard deviation of timing between them is ~0.22 seconds, which is fairly high.
That level of timing variation suggests the audio is not heavily quantized and was likely played by a human, or at least simulates human timing with some natural feel. A tightly quantized track would typically show much lower timing variance (e.g., <0.05s).
Of course, it is an LLM. Loads of salt 🧂needed to accept its claim.