Ask HN: Linux-Compat Hardware for Self-Hosted AI Models & gen advice

Hello, I'm trying to enter the realm of self-hosted AI (maybe, eg llama3.1-8B) and would appreciate suggestions on both recommended 'budget minimum' hardware and specific Linux distros if any. For hardware, I'm assuming at least 16-32 GB ram, but years ago when I first entered Linux, there were lots of issues with non-intel cpu/gpus, so I want to make the best decision here, and avoid unnecessary upgrades too.

Considerations:

I will not be generating or working with images. If this changes, I'll use an remote service.

I will eventually want to train a model or optimize/tailor it towards my needs, which are not yet set, but have been mostly research and code generation.

I'm about as ignorant on the general subject as your average grandma and my python is withered and was feeble at its best. Also, while not necessarily a big deal, the necessity of using pip is slightly awkward in Debian, requiring the use of venv. Is this standard, or are there distros optimized for easier integration of repos and Python?

My aim here is not development, as that is legitimately beyond my potential. Rather, I hope to eventually become reasonably surface-knowledgeable on the subject and formidable as a user. Recommended resources for proceeding from the very beginning, as a grandma, would be helpful. My comfort level in Linux is pretty high, but I gave up on understanding the inner workings when systemd was introduced and have become a bit of a thug, just bullying my way around and making things work often without any deep understanding. That's basically where I am. For anyone that does reply, feel free to mention anything you think relevant, eg misconceptions exhibited by my text here, or whatever.

Thanks, if applicable.