use of ChatGPT for HIFI

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.

unfortunately chatbots are nowhere near reliable enough to trust answers to even trivial questions.  you end up having to do a regular search to view the source material directly.   and even when you correct it, the chatbot will eventually repeat the same errors.  i was excited to use one at first, but rarely do any longer.

be warned, they are designed to stroke your ego and lure you into using them more and more.  that is their real purpose, not providing accurate information.

I recently used both ChatGPT and Grok to help me narrow down which amp and speakers I should keep as I’m considering downsizing. I queried 4 amps and 4 pairs of speakers: Luxman LX-380, Luxman 550axMarkII, Audio Research i50, Musical Fidelity 6si and potentially a LTA Z40+. Speakers: Dynaudio Heritage Specials, Harbeth30.2XD, Devore Super Nines and Magnapan LRS. I added room dimensions etc, types of music I like, % streaming vs CD and vinyl and whether I listen when doing things around the house or stationary. I asked to list  the pairing 1-2-3. 
The results were quite similar between the two platforms and both gave very detailed summaries of their results. E.g. why the pairings make sense  based on a number of factors. Anyone care to guess what I’m keeping?

I recently started using ChatGPT for comparisons. Sometimes the information is just wrong. When I pointed this out, it adjusted and provided different results. It is just like any other resource at this point. It is just retrieving information from the internet and re-packaging it, albeit a bit more efficiently than if I was pulling all the information together myself. I use it as a starting point. Ultimately, we need to determine what sounds best in our particular systems, unless you want to jump into the ASR pool and rely only on data points rather than what brings joy to you. I still trust my ears and judgement.

I use perplexity and asked it to analyze my system.
perplexity did a good job of providing a lot of detail about my system, and how it could be expected to perform. It did make a few suggestions, but they seemed pretty generic to me. It almost seems like a super search engine that rapidly assimilates what’s available on the net into a reasonably coherent package. This is obviously just a rudimentary AI tool, and I consider it useful, but not really intelligent, more of a compiler.