When discussing privacy, a crucial question arises: what's your comfort level with exposure? Are you posting via Tor, or are you more relaxed about your online footprint?

The truth is, everyone has a different threat model. To avoid confusion, you should be clear throughout your discussions about what your threat model is. Some users face high-stakes surveillance in oppressive regimes, while others are simply ad-averse and not concerned with anonymity. Then there are those who prioritize absolute privacy no matter how small the data leakage is.

In reality, privacy doesn't have to be overwhelmingly complex. However, things get murky when you have a weird mix of yielding data to multiple, potentially untrustworthy, data collectors – including corporations and governments. Consider your daily habits: do you regularly use a credit card, or activate your phone's cellular connectivity? While many are comfortable surrendering their purchase records (credit card) and location history (credit card + phone) to data collectors (which often retain this information indefinitely), others bristle at the idea of relinquishing so much personal data without discernible benefits.

Interestingly, Jukebox from OpenAI was trained on what appears to be copyrighted music and involved styles and renditions that explicitly referenced specific artists. It's now four years old though. The demo songs don't seem to be available anymore on Soundcloud.

There is MusicLM from Google (2023) - no lyrics. Also, AudioCraft from Meta (2023) - also no lyrics as far as I can tell.

Llama 3 70b is almost as good. Self-host it and sleep well at night.