I built and trained a GPT on OpenAI's 4.o using Ivy League Master's courseware, as well as consulting back and forth about what it did and did not know, ultimately rolling the info back into the GPT's base training. I gave it my detailed system info and while it is many times better than when I started, it still struggles with some nuanced hallucinations, mostly resulting in it agreeing with me when it omits considerations (at least until I tell it what it did and to fix it).
I recently had a great conversation about my evaluation of a preamp replacement... moving from a decently modified Schiit Freya + to either a Benchmark LA4 or a Mark Levinson No. 5206. The objectives were to provide better line matching voltages and to keep the coloring of the buffering / follower stage of the sound to match the 1947 Telefunken AT12. It recommended the LA4 and also provided step by step instructions for how to match the Schiit Tyr input voltages.
FWIW - I have a separate phono preamp I am not looking to upgrade.