Using a spare system that isn’t in active use anymore to mess around and learn is definitely a good idea. However for long-term use I would keep an eye out for energy consumption. Depending on electricity prices it might make more sense to get a newer system with lower power draw, especially for something with high uptime.
You and all the companies training their AI models, like Meta for example
My understanding would be:
Probabilities: Chance of a certain outcome happening. E.g. Outcome A has a 70% chance B a 20% and C 10%
Possibilities: what outcome scenarios exist. E.g. there exist 3 (A B C). Those Possibilities might have a probability associated with them
Plausibility: looks at the degree of truth of a statement. So if it logically makes sense and is the correct answer/is what happened. You might make a judgement of the plausibility of a possibility based on the probability of it happening. Say if something has two outcomes one with a 99% chance and the other 1% then that might be the more plausible one. Or if it has no chance, then it might be implausible
Edit: since someone mentioned the example of a coin toss.
Head and tails have a probability of 50% each (for the sake of simplicity I assume it won’t land standing up)
A coin toss has two possible outcomes (possibilities). Head and tails.
Someone says he flipped a coin and got head 1000 times in a row. That’s not plausible with a fair coin because of the low chance of it actually happening (even if there is a indefinitely small chance). As a result you might assume he is either lying or the coin is weighted for that outcome.
I think for businesses that deal with physical items, rather than offering services, and are not drop shipping, selling something 3D printed has to be one of the lowest inventory options. You need to have the printers, some filament, and maybe some accessories like packaging boxes, but otherwise you can just make your product whenever you get orders (or have a small stock for faster fulfillment).