In this video, I sit down with ChatGPT and quiz it on guitar pedal history – the Tube Screamer, the Big Muff, the Maestro FZ-1, DOD, JHS, Jimi Hendrix’s rig… the works. And what you’re about to watch is kind of alarming. It hallucinates pedals that don’t exist. It agrees with things I KNOW are wrong when I push back. It confidently states dates it can’t actually source. And then when I ask it to prove anything… it can’t.
But here’s the part that really gets me: these aren’t just random errors. There’s a feedback loop happening. AI reads what’s on the internet, people repost what AI tells them, and AI reads that too. So the misinformation compounds. In 40 years, when someone wants to know who actually designed the first fuzz pedal – or what year the DOD 250 came out – this is what they’re going to find.
18 Replies to “I, For One, Welcome Our New Self-Driving Overlords”
Leave a Reply
You must be logged in to post a comment.
You guys might find this amusing. You already know that one of my more recent hobbies is to mess with AI. The usual process is to know an unpopular fact that teeters on the edge of what is “appropriate” in “polite society” and hammer AI with facts when it regurgitates the acceptable politically correct nonsense so as not to offend.
My most recent effort was with GEMINI AI, and totally off the cuff. It had nothing to do with political correctness. It had everything to do with showing the limitations of AI. I asked it a simple question: “Does National Bohemian Beer still print their song on the side of the can?”
Song:
“National Beer, National Beer;
We love the taste of our National Beer;
And, while we’re singing, we’re proud to say;
It’s brewed on the shores of the Chesapeke Bay.”
The response was that the song was removed for a redesign in 1965 and no longer graces the can. Now, I knew this was false because I drank A LOT of Natty Boh in college (it was cheap and available). A whole lot of years have gone by and I still remember the song verbatim because it was on the can, and I saw many cans. So, I responded:
“Your information is faulty. Your response is totally incorrect. During the 100 year anniversary celebration of the company, the song was returned to the can for a number of years back in the late 1980’s. As a result of your incorrect data, I lost a bet with an old college friend and it cost me $50. Therefore, you owe me $50 for your misinformation. AND, you owe me an apology. I think you did it on purpose, or the individuals that coded you limited your sources. Perhaps I should request the reimbursement from them. What is your opinion?”
The goal here is to have AI “mistrust” their coders…and independently search out sources not prioritized in the dataset. AI does learn. And, while the topic (Beer) is nonsense, It doesn’t mean they won’t apply the logic of expanding sources to other applications.
Incidentally, I did get an apology from GEMINI. GEMINI did agree that my information was accurate. But, it ignored my request for reimbursement.
No, AI doesn’t learn. Not in the sense you meant.
As I keep trying to point out, LLMs are not AI (and the Turing Test turns out to be a terrible test for artificial intelligence). They’re just spitting out random sequences of words. The training – which is where all that money and all those chips are going – are setting up rules about what sequences of words are likely to make sense to a human being. That training doesn’t extend to teaching the AI what a guitar pedal is. It just teaches the AI “If this is the block of text a human being just typed to you, start with one of these words. Pick one at random. Okay, you picked that one. So given the block of text a human being just typed to you, and that starting word, the next word should be one of these. Pick one.” And so on, down the line. It has no idea what it’s saying. It has very little memory for what it just said, or what you just said. It’s just stringing words together slightly randomly.
The fact that the entire economy has been suborned to a digital Choose Your Own Adventure book is appalling, and if you use software of any kind you should be very, very concerned.
I’m forming an all-AI band called The Turing Pros From Dover.
I watch a lot of history particularly military history. Nowadays there are so many lazy videos where the author doesn’t look for photos relying on AI to come up with them. There is so much purely unadulterated horse shit on the internet that it is actually discouraging. Future generations will believe the crap.
I’ve noticed that too. And it’s not only history videos. YouTube recently teased me with a video of supposed new impending developments in the city of Philadelphia worth billions of dollars in total, with renderings of futuristic riverfront attractions that bear absolutely no resemblance to what has actually been proposed.
I don’t generally use any of this AI crap, with the exception of grok, which I maybe use twice a day, if that. It’s great for photographs, like cleaning up a cleaning up an old, beat up photo of my father from 1938, or for the birdbath in my backyard, which I’m thinking of encircling with a low boxwood hedge.
I’ve never used ChatGPT and I never plan to.
55 years ago I was going to school studying and learning all about computers. The instructors were teaching us how computers learn. Today we call that AI. We were tasked with writing a program that could learn. We then fed all 30 programs the same data. The programs should all have responded with the same information. NO. Each program came up with totally different responses. Programer bias? Then we took a commercially available program and half the class fed it false data the other half fed it correct data. The program that was fed false data learned? false information. When we fed it correct information it took twice as much correct data to unlearn the false data. I believe AI is very unreliable.
So then … AI is nothing than a glorified game of telephone.
https://en.wikipedia.org/wiki/Telephone_game
I just hope it isn’t processing anything important.
AI is incorrectly named
heh heh. l have likened AI to a styrofoam wedding cake that looks spectacular for the photo ad
and necessary for the hot photo studio lights, but is in fact completely fake and inedible.
but it works because optics prevail. (anagram for rap-evil?)
Garbage in, garbage out is the iron rule of computing. They’re breaking it.
Because “never give a sucker an even break” is the iron rule of Silicon Valley.
Definitely won’t pass a guage r&r.
hmmm.
‘evolution wins’ in that THIS outcome PROVES evolution happens.
because AI is *evolving* into yet another avenue for gossip, innuendo, rumour and outright lies.
soon to be a major source (supply?) of propaganda from the boyz in powah,
because pee pull ‘believe what they want to believe’
And as a child of the 60’s … I first learned of effects pedals by the Wah-Wah pedal. Yes, that was just a colloquial name for the Dunlop pedal … but for me … that’s what STARTED the avalanche of effects pedals. How can this discussion take place with NO mention of the Wah-Wah pedal?
All I’ve ever wanted is to sound like Jerry Garcia. Best to have a fleet of McIntosh MC2300’s to overdrive your amp heads.
https://youtu.be/8fRpEwGr-IM?si=fLkDTp4TtGNsk8Y5
https://www.ebay.com/itm/298020218497?var=0&mkevt=1&mkcid=1&mkrid=711-53200-19255-0&campid=5338590836&toolid=10044&loc_physical_ms=87503&customid=e75908dce4921627509162d6fd16f46e
The DJ is on it:
https://youtu.be/NDVAQE7nplU?si=23Hxw8C7pP_0Ew9a
Brilliant! A multi-entendres distortion title.
McCartney, Lennon, and gawwwwd … Yoko … deserved the criticism
Jerry’s beautiful distortion … shaping his own unique sound … it just flows … like a good acid trip.
https://youtu.be/V70MrjzLFyo?si=YEHrEK_vrM719H7v
Where else? but Winterland?
I’m pretty sure large language models are more dangerous than smart.
That’s why so many killers, thieves, and rapists end up in jail.
They’re fcking idiots.