Startling AI results.

In the words of a 1928 New Yorker cartoon “I say it’s spinach and I say the hell with it.”

Joking aside, I’ve been asking “is a better than b” and the ai answers often are incorporating what I believe to be marketing hype—either from the manufacturer’s website or more likely from a bad you tube reviewer parroting what he’s been told. In any case, it’s not quite trustworthy but it’s sorta trustworthy.

Ai does better with more general questions like “which technology has advanced more in the last five years, streamers or DACs. 
 

 

As for being exacting in word choice, the burden is on the technology.  Way too many variable in people’s language skills. In fact, one the the benefits of AI is that I don’t have to form a complete sentence to get a reasonable response. Years ago, I spent quite a few years as a reference librarian, both in public libraries and government R&D centers, and believe me, queries come in all shapes and forms and I actually think AI beats the the human is sussing out the real question.

I once spent a few minutes explaining that I could not identify a book titled  “Homeless Idiot.” When the patron insisted it was a classic and her son needed it for school the lightbulb went off: Homer’s Iliad. 

There is also the concept of prompting for great prompts.  Interview the AI agent and ask it what questions you need to answer for it to best answer your question.  Something like “If I were to ask you to find the weak link in my audio system what questions would you ask me to build a prompt to help you answer the question”.  I got this set of questions in a templates format from Gemini.

Copy and paste this template into your prompt, filling in as much detail as possible. The more blanks you complete, the more precise the diagnostic will be.

1. The Hardware Ecosystem

Speakers (Make & Model): [___]

Subwoofer(s) (Make, Model, & Quantity): [___]

Amplification (Integrated or Separates, Make & Model): [___]

Digital Source (Streamer, Transport, DAC): [___]

Analog Source (Turntable, Cartridge, Phono Stage): [___]

2. The Room and Positioning

Room Dimensions (L x W x H): [___]

Room Layout (Closed room, open-concept, L-shaped, etc.): [___]

Speaker Placement (Distance from front/side walls, distance apart): [___]

Listening Position (Distance from speakers): [___]

General Acoustics (Reflective/Hard vs. Damped/Soft): [___]

Treatments & DSP (Acoustic panels, bass traps, room correction software): [___]

3. Infrastructure

Interconnects (Make/Model, RCA vs. XLR): [___]

Speaker Cables (Make/Model): [___]

Power Delivery (Standard wall outlet, power conditioner, dedicated circuit): [___]

4. The Subjective Gap

System Strengths (What it already does well): [___]

The Symptoms (What feels constrained, e.g., muddy bass, harsh treble, flat soundstage): [___]

Reference Material (Specific tracks/genres where the system struggles): [___]

Primary Goal (What you want to achieve with the next upgrade): [___]

AI is like talking to a 6 year old all it knows is wht it is told or the parameters that it is given. Ask all of them or just 2 the same question and you will get different answers. Ai might be great for computing but for everyday stuff it's a big miss....

I’m not sure what percentage of folks are using free version of AI vs paid services.  There is a world of difference between free and paid.   Exponentially.   Free just touches a subject.  Paid digs deep, reasons and much much more.  I use paid versions of ChatGPT, Gemini and Claude.   Not going back to free.  Makes no sense to go back to the Stone Age. May drop one at some point.  You can feed it lots of variables with one request and or comments. Comes back a fair amount of info and from there continue the chat and go further into the subject.   If you are on a long chat focused around one subject, and you’re ready to wrap up the chat, tell it to give you a summary in table format, produce a pdf file using color. Sums it nice and tidy.  And thats using AI lightly.