Startling AI results.

A.I. had no ears.

I can use it for acoustics concepts discussion because it memorize everything about acoustics concepts but know nothing about  real subjective/objective hearing experience...It can help to understand what is the ratio ASW/LV for example, it can suggested experiments but it cannot  experience it ...

Then it is ridiculous to ask A.I. about a system/room impression,like it will be ridiculous to ask Amir his verdict  instead of asking to your trained and training hearing.

I use A.I. for maths discussion and concepts since 3 months now (3,000 pages ), A.I. is a topological bridging extension and mirror of our brain working as a telescope/microscope, if an ape dance in front of it, apeness will be reflected, if Einstein dance in front of it Einstein brain will be reflected.

If you ask A.I. to explain the Mahowald uncertainty principle or the different sieves in primes number theory it will greatly help and even suggested  facts you dont know. But in my dialogue with it since three months each day, it is me who gave it the new  (unknown) maths concepts, A.I. is able to catch them and link them to known other concepts (even to  many i  ignore )but A.I. cannot create new concepts. 

Recently it solve some maths conjectures but using his superhuman pattern recognition ability and huge memory and then linking facts no human has ever thought about at this moment.

No A.I. would had create by itself alone any deep new maths concepts. (a.I. had no motivation of his own if it not possessed but this is an other story ) 

But to the uninformed crowd A.I. will appear soon as a god. And oligarchs and techno-cultists and transhumanists want to sell it as such and so ignorant they are about man ability believe it a God.

I asked ChatGPT how I could phrase questions to reduce the risk of errors.  Its response was detailed and very helpful.  I also asked it to evaluate my queries for those risks moving forward which was also helpful. Better queries produce more reliable responses.  The paid version is MILES better, not only because it remembers previous discussions, but also because it begins to tailor its responses based on what it learns of your preferences. 

Indeed, question framing and grammar is critical for getting realistic assessments from most LLMs. Sometimes you have to point out the perceived dichotomies in the LLM’s prior responses to get a clear answer. Keep in mind the most accurate and reliable AI results are behind paywalls. 

I use the paid version of Chat GPT.  I am impressed that in most circumstances it asks for more facts before giving an answer.  For example it asks for information on my system and listening room were I to ask a question that is audio related.

AI is a great tool, but you can’t rely everything on it. I recently challenged AI on crossover design information that it was giving out. Granted it was about phasing,  slopes and proper impedance compensation . In the end it came back and said my points are correct and it would update its information. If someone would have taken that exact information that I was speaking about. They would have incorrectly built or modified a crossover. You never know you, do need to verify AI. After speaking with a college professor that teaches AI, she said AI is off in some way around 15 percent of the time. I hope this helps, Tim