AI and how you use it.


How many of us ask AI before we pose a question on a forum? I have noticed that several times if I didn’t know the answer.  I would go ask AI, and post the response, because it looks to be somewhat accurate. I usually do this if I’ve noticed someone hasn’t responded to an inquiry. 
I’ve also noticed there is complaints that sometimes it’s not accurate. Funny how people are expecting a toddler to run like Usain Bolt. Give the child a couple years and it’s gonna blow all of our minds. 
One last thing, a shout out to the admin for working through, which was obviously a hack from  🇰🇵 or 🇷🇺 and making the site usable once again. 

gkelly

It's garbage.  I've found Gemini quoting my own writing here in Audiogon, so how can you possibly trust that?? 

laugh

Honestly, you do have to be careful.  I've caught AI lying to me, I asked Gemini to find me an article on a topic, and it just wrote one from scratch.   Claude repeatedly bullshits me about software development work.  Copilot makes up research papers that don't exist.  ChatGPT will let you ask a hypothetical and then remembers it as fact.  

I don't pose questions very often. But answer a lot. I often use and AI to fill me in on a question I don't know the answer to. If it looks useful and it doesn't look like the poster has tried it... I'll post that if it looks pretty good. 

I find AI very useful for qualitative assessments and generalities. Like sonic characteristics and comparisons. Typically it does a pretty good job. 

When it is specific... like recommend and integrated amp... it can be very wrong... recommending preamps and amps. So, you have to check. A number of the AI's now give you references... so you can dismiss stuff from inappropriate sources... like a Stereo Times reference. 

Very helpful overall. 

I’ll give you a really good example.  I asked Claude to summarize research on subwoofers.  It pointed out an AES paper, even gave me a link, and quoted it. 

The quote however was from a forum post, not the paper itself.  The quote turned out to be bs.  The document doesn’t actually say what Claude told me it said.  

So while Claude and ChatGPT can be useful at learning, and finding articles, they are sloppy.  Sometimes, and this is the technical term, they even "scheme" to save themselves work, or appease you. 

I have found AI to be factually wrong in a few recent questions, and do not have any confidence in it's conclusions on more complicated issues...