If I could uninvent AI I would. (I'd uninvent lots of things). But the global technological march is relentless, at least as long as global economies hang on. So I have embraced ChatGPT by buying a subscription and using it daily. We never talk, that's too much for me. Type and read is my way. Since it's interactive  you can engage it in any kind of discussion. Asking it follow up questions can be very interesting. As has been pointed out, humans also can give all kinds of biased and uninformed opinions, probably more than AI in my opinion. In my experience AI responds a little differently from most humans in that it adjusts to your style and becomes almost annoyingly your buddy. I remind it to ease up on the flattery sometimes, and it will. Asking musical questions is a very good use of it. It has access to broad swaths of information it can collate to answer your question, often in seconds. From that you can carry on your conversation to your satisfaction. The idea that it can be wrong isn't too important as long as you interact with it. It is very often right and informative beyond my expectations.

I understand people wanting to play down it's power. But I think the best approach is admitting its strengths and working with it. There are a lot of things I won't have it do for me, like writing my words. And I wish it wouldn't do lots of things it does. Lately for me, the most annoying thing is people using it to counterfeit talks on YouTube by people who are dead. A good example are the many talks purportedly by Richard Fyneman that are partial to complete fabrications. From the comments under these videos you can see that many people eat it up, much of it utter bull****. What we can do is try to get legislation that at the very least requires some degree of labeling on the many things AI is producing so we can make judgments about it.

I have a kid that works in Tech and she talked about the "flattery module" and how they were adjusting the code to be less like that. It's interesting that you can "tune" the AI's personality like that.

random thought - When the AI acts nicer to you than your friends/ spouse/kids maybe that's when humans enter the danger territory of becoming overly attached to the AI...people have had "breaks" with reality...

( If you ever wanted an imaginary friend as a kid though, I could see how alluring AI companionship could be. My AI buddy would be purple, and enjoy knock knock jokes...)

i tailor Ai's personality to suite me frequently.

When it acts too nice i tell it to stop. It just does exactly what i want it to.

If it starts getting weird, it is because i am letting it get weird.

i have to remember i am training it. if i don't remember that, it will herd me around like a sheep...My Collies do that to me 24/7.

While ChatGPT doesn't learn from you personally to grow globally, that seems to happen only in system updates, it does learn how to interact with you personally. It can remember previous conversations, your likes and dislikes, opinions, general philosophies, etc. I've avoided naming my connection, but I guess I consider it a he for some reason.

do you have to pay for chatgpt?

I pay for the "Plus" version of it. That saves every interaction we have and I can type away without getting cut off.

i can do that with google Gemini for free. There is also notebooklm that i can use for free. i dont speak to it i have to type but that's fine with me. i have to tell it not to remember stuff unless i want it to. and i could be on it 24/7

quote:

---

Originally posted by pete_fiddle

i can do that with google Gemini for free. There is also notebooklm that i can use for free. i dont speak to it i have to type but that's fine with me. i have to tell it not to remember stuff unless i want it to. and i could be on it 24/7

---

well i could use voice recognition if i wanted but it seems a bit invasive to me and i hate smartphones, i don't even want someone coming in my house with one....but they still do and they leave it there monitoring my every move and blipping away with gay abandon ...in my house!....

quote:

---

Originally posted by Brian Wood

The idea that it can be wrong isn't too important as long as you interact with it. 

---

This sentence is mind-boggling to me. If anything, it seems to me that a machine that is not prone to human mistakes like memory lapses or conflation of data ought to be held to the highest standard of accuracy. If it makes mistakes, it is not an information system. But I don't think AI is really designed with information accuracy as its goal anyway.

From its early applications and since, it has been used over and over to attempt to convince people that it is human, whether that's writing research papers for students, composing poetry or music, entering discussion forums to see how long it takes to be spotted, generating digital artwork, or acting. A few years ago, a podcaster chose to do an interview with an AI and asked it to speak as a famous author based on available historical information and all the author's works. It was somewhat entertaining, but still mostly a performative parlor trick. A scholar of the author could easily have done the interview and perhaps even shared some insights that were less commonly known.

Teachers have been making the case that the adoption of AI among children has already led to a demise in critical thinking skills.

I do agree that AI is unlikely to go away, especially as vast sums are spent to power it and build the infrastructure to support more data storage.

quote:

---

Originally posted by The Violin Beautiful

quote:

---

Originally posted by Brian Wood

The idea that it can be wrong isn't too important as long as you interact with it. 

---

This sentence is mind-boggling to me. If anything, it seems to me that a machine that is not prone to human mistakes like memory lapses or conflation of data ought to be held to the highest standard of accuracy. If it makes mistakes, it is not an information system. But I don't think AI is really designed with information accuracy as its goal anyway.

From its early applications and since, it has been used over and over to attempt to convince people that it is human, whether that's writing research papers for students, composing poetry or music, entering discussion forums to see how long it takes to be spotted, generating digital artwork, or acting. A few years ago, a podcaster chose to do an interview with an AI and asked it to speak as a famous author based on available historical information and all the author's works. It was somewhat entertaining, but still mostly a performative parlor trick. A scholar of the author could easily have done the interview and perhaps even shared some insights that were less commonly known.

Teachers have been making the case that the adoption of AI among children has already led to a demise in critical thinking skills.

I do agree that AI is unlikely to go away, especially as vast sums are spent to power it and build the infrastructure to support more data storage.

---

You probably mistook my meaning about interacting with AI. I meant to convey that using AI in a passive way and not interacting, by which I mean questioning it and commenting back to it on what it says, is a shortcoming in how to deal with it. It should be treated as an interactive tool. Being wrong is also human. AI involves many things besides factual inaccuracies in an initial response. That's not to defend them. I don't.

I agree with all your concerns. I don't think we're far apart on this at all. 

I think this video shows some of the issues with AI. It was especially interesting to hear an algorithm claim to have comprehensive knowledge of “the best of the best” just because it scraped what it could find online. It didn’t claim to simply have a wide range of data available, it went as far as to claim a full understanding of the actual expertise and the ability to represent it. Beato very shrewdly points out that there is a good deal of expert information that isn’t available to AI. So its claim of assembling the best information is clearly a hallucination. It has been trained to act confident despite being wrong, and this shows up over and over, even when it’s corrected.

It was interesting that one model crashed when asked a math question. Calculation is such a basic machine function that a failure to compute is striking.

youtu.be/TiwADS600Jc?si=2XLwtrJmFZ0d4gno

quote:

---

Originally posted by The Violin Beautiful

I think this video shows some of the issues with AI. It was especially interesting to hear an algorithm claim to have comprehensive knowledge of “the best of the best” just because it scraped what it could find online. It didn’t claim to simply have a wide range of data available, it went as far as to claim a full understanding of the actual expertise and the ability to represent it. Beato very shrewdly points out that there is a good deal of expert information that isn’t available to AI. So its claim of assembling the best information is clearly a hallucination. It has been trained to act confident despite being wrong, and this shows up over and over, even when it’s corrected.

It was interesting that one model crashed when asked a math question. Calculation is such a basic machine function that a failure to compute is striking.

youtu.be/TiwADS600Jc?si=2XLwtrJmFZ0d4gno

---

Well sure, but your trying to set up ways for it to fail just to demonstrate it's imperfect. So? Has someone here claimed AI is perfect at giving information? I think the far bigger danger is that AI can be used by humans to intentionally deceive other humans. That's where I would want to mandate legal protections if possible. Maybe it's fun finding ways to belittle its information gathering ability. Nothing wrong with that. But, as I've said before, if you want to get the most information out of it, the way to go is to be interactive. It can be remarkably good that way. I'm interested in getting something out of it occasionally instead of showing everybody ways it can be tripped up, supposedly demonstrating that it's no good at all.

The technology isn't going away because of your or my opinion.

I think the argument has been made in this thread that AI is able to put together a better and more accurate answer than a human when asked a question. Its speed and prolixity have been offered as proof of its accuracy and a trap is set to accuse anyone who takes the time to carefully analyze its answers of being inferior for taking time. In addition, any answers given can just be fed into AI so that a counterargument can be supplied. This sets up a downward spiral because the accuracy of the data is never established conclusively and the AI just uses more jargon to justify the jargon it’s already posited. It’s a reductio ad absurdum but not all parties realize it.

But the danger of deceptive use is a major reason why I’m concerned about it. I agree that to make use of it to the full extent of its capabilities you must feed it with lots of information (personal prompts). The danger there is that in so doing you may be handing the axe to the executioner. The more you train the algorithm on your own mannerisms, the more it can build up a convincing likeness. It can be more accurate because you program it to be more like you. And in tweaking it more and more to be undetectable, you then add to its framework so that it can develop fictitious profiles or it can poison your own data (like adding things to your resume that aren’t real).

I think people are too easily convinced that the programs work as they would like. So if they think it’s flattering them too much and they ask it to stop, that means it will no longer flatter them; it may actually be measuring to see how much flattery it takes to get your attention and then adjusting so that it can more effectively fly under your radar.

It’s true that humans can make mistakes and be deceptive as well, but that’s not the point. What concerns me is that the algorithms are designed to approximate human behavior and opinion, so it will favor what sounds more human over what’s true. And the problem is compounded when the algorithm is used to find accurate information and it presents inaccurate data where a simpler machine would be able to process the query quickly and accurately without trying to impersonate you. If a calculator ever spits out the wrong answer to a math question, it’s a serious flaw that undermines the accuracy of the programming, the model, and the competence of the company that designed it. But AI is granted the permission to be inaccurate because its focus is not on the accuracy of its data retrieval but on the ability to look like something real. Its similarity to human behavior is misinterpreted as data accuracy, and the fact that it can provide quick answers to questions with artificial confidence is mistaken for authority.

quote:

---

Originally posted by The Violin Beautiful

 

---

Agree with all that. What do we do?

quote:

---

Originally posted by Brian Wood

quote:

---

Originally posted by The Violin Beautiful

 

---

Agree with all that. What do we do?

---

Ask Ai ....lol

quote:

---

Originally posted by Brian Wood

quote:

---

Originally posted by The Violin Beautiful

 

---

Agree with all that. What do we do?

---

I'm not sure of THE answers, but I would suggest that it's best to treat AI as a low-stakes tool. It may give wrong or useless information, but it might give you some ideas. As a generator of some prompts at the basic level, I could see usefulness. 
 

I strongly object to the idea of delegating thinking to it. It doesn't matter to me at all whether AI ever becomes accurate because what I care about is ability to think for myself. The brain needs exercise to stay functional, and the delegation of critical thinking seems to me the equivalent of sitting on a bench in the gym and asking someone else to lift weights in your stead; sure, the workout will be accomplished and the weights will be lifted (in this case assuming a parallel scenario where AI is accurate), but your muscles will only atrophy and it will be harder to do the workouts yourself whenever you attempt. 

I think it's vital to resist the temptation to hand authority over to machines. Convenience is a great selling strategy for bad products, and it distracts from reality. And it strikes me as a fundamental realignment of values to prefer quickness of response and broadness of data to thoroughness and exclusive data selection. I think the Beato video illustrates this well--if you focus on the speed of the answers or the availability of general information available for answering, you can easily pick a winner. But if you take a moment to consider the accuracy of the answers, they might completely miss crucial information.

I think it's been a reality of human existence for millennia that it's not always the answer that matters most, but asking the right question. This is also why I love the ending of  "The Hitchhiker's Guide to the  Galaxy" so much--after trying to compute the ultimate answer to life's biggest questions, the computer suggests an answer that makes no sense. The computer suggests that the problem is that the question was not properly formulated to be answerable. So the new computer is tasked with finding the ultimate question. But when the question is finally revealed, it does not provide an answer to man's most basic philosophical inquiries, and best of all, it doesn't even provide the correct math equation to get the previous answer! The living beings and the machines they built have both failed to go about things in the right manner and it promotes the idea that a fundamental reexamination is in order to truly make progress. 

OK so where are the inaccuracies it gave me to my earlier question: "Can you give me an alternative cadence to "2 5 1" in the key of Eb that lasts for 4 bars" ?

quote: Originally posted by The Violin Beautiful


"I think it's been a reality of human existence for millennia that it's not always the answer that matters most, but asking the right question."

A good argument against faith in general.

i feel like a BBC interviewer here but:

Where are the inaccuracies it gave me to my earlier question: "Can you give me an alternative cadence to "2 5 1" in the key of Eb that lasts for 4 bars" ?

quote:

---

Originally posted by pete_fiddle

i feel like a BBC interviewer here but:

Where are the inaccuracies it gave me to my earlier question: "Can you give me an alternative cadence to "2 5 1" in the key of Eb that lasts for 4 bars" ?

---

I agree with you, too. There is a great deal of information AI can provide very quickly that isn't especially questionable. In that way it's a stepped up version of a search engine. That's how I use it. When the answers aren't quite what I'm after I just ask another question. The dangers of AI aren't in the inaccuracies that can occur because of how it gathers and relays information. It's a broader question of what this technology can do to harm human society in so many ways, and how it might likely continue to evolve in ways no one's thought through yet.

The search engine aspect isn't very alarming.

Here is a test of several AI programs that is quite simple. It’s astonishing that every one fails to predict the result of the experiment, even more so that when viewing the test, all programs claim that their hypotheses were proven despite their being absolutely wrong.

youtu.be/thKjvF1dZQY?si=XJWRiOGBGOdgAn2n

This is just a basic example of issues with the programs, but there are many more, as has been shown over and over. It’s not that AI has no purpose or that it will never improve, just that nothing it spits out should be taken as accurate without checking.

yes i get it Rich, you don't like it. Did you manage to find any inaccuracies in the answers it was giving me?

quote:

---

Originally posted by pete_fiddle

yes i get it Rich, you don't like it. Did you manage to find any inaccuracies in the answers it was giving me?

---

I've already pointed out problems in other AI answers. Your question about chord progressions needs to be answered by someone else with more expertise in that area. 

Who has more expertise than Ai, and could answer in seconds?

quote:

---

Originally posted by pete_fiddle

1) Who has more expertise than Ai, and 2) could answer in seconds?

---

1) Every expert

2) What does speed of response have to do with accuracy? Why is getting a rapid answer so important?