good enough simulations that you can’t tell the difference.
This requires us having actual conversations with those dead people to compare against, which we obviously can’t do.
There is simply not enough information to train a model on of a dead person to create a comprehensive model of how they would respond in arbitrary conversations. You may be able to train with some depth in their field of expertise, but the whole point is to talk about things which they have no experience with, or at least, things which weren’t known then.
So sure, maybe we get a model that makes you think you’re talking to them, but that’s no different than just having a dream or an acid trip where you’re chatting with Einstein.
I switched from raspberry pi and orange pi to a cheap Intel NUC, and I think it’s just a much nicer experience.
The pi is great fun, but the HW transcoding on a NUC “just works,” and the SSD and 16GB RAM opens a lot of doors. My N100 NUC was less than $150, and it included everything (case, power supply, 500GB SSD).
My pi found new life as an off-site backup: attach a big HDD, set up WireGuard, and have a cronjob do daily rsync and snapshots. I have it set up at in-laws, and it works great.