Latest / Connor with Honor: AI, Real Estate and Real Life / Don't limit AI by your limited prompting abilities

Transcript

Don't limit AI by your limited prompting abilities · Connor with Honor: AI, Real Estate and Real Life

• 0:00
  SPEAKER_00: Maybe it's not exactly funny, but you do see that people are building this technology that's really, really smart.
• 0:07
  SPEAKER_00: And there's probably nothing wrong with that, because in the technology, in artificial intelligence, we can cure things.
• 0:13
  SPEAKER_00: We can solve mathematics.
• 0:15
  SPEAKER_00: Now, the mathematicians probably don't appreciate that, but of course, it can be solved, and it looks like probably all of it will.
• 0:23
  SPEAKER_00: And in fact, with physics and mathematics, there'll probably be new things discovered in very short order because AI, while it maybe used to be the next word predictor in the way that it works, now it's coming up with its own ideas, its own thoughts, its own processes, its own conjecture, its own understanding of things.
• 0:42
  SPEAKER_00: And then it's offering things we've never seen before.
• 0:45
  SPEAKER_00: You can just track it through the different games that it's learned how to play and how it's applied solutions that us human beings have never even considered, even though some of these games have been around thousands of years.
• 0:57
  SPEAKER_00: A good example is Go, do some research on it.
• 1:00
  SPEAKER_00: G-O, it's the name of the game, if you're not familiar.
• 1:03
  SPEAKER_00: So when people say that this is just a repetitive type of computer that's just guessing the next word because it has access to all the data, has a lot of power and a lot of ways to compute that particular data, it's not that any longer.
• 1:17
  SPEAKER_00: So it has really transitioned.
• 1:19
  SPEAKER_00: If you start to use ChatGPT or Claude or these other systems, really, those are kind of the two Gemini as well.
• 1:25
  SPEAKER_00: So maybe three are kind of running the game at the top, you'll see that they do a lot of work and they're able to accomplish a lot of your daily grind, your tasks.
• 1:36
  SPEAKER_00: And the more that it's connected to online systems or the internet or whatever it may be, your browser, whatever your work platform happens to be.
• 1:45
  SPEAKER_00: Maybe you're an Excel person, maybe you use Word, maybe you use whatever it may be, these uh different Google Tools uh sheets or whatever.
• 1:53
  SPEAKER_00: And it goes into those, it's able to manipulate them, input data, extract data, change things, modify things, of course, by your command.
• 2:01
  SPEAKER_00: And in some cases, the different, depending on the amount of permissions, the more you give it, the more it's able to do.
• 2:06
  SPEAKER_00: In real estate and in business, a couple different things that people are often talking about, they're still using different structures like different components, like Canva, for example, and they're using the AI that's a part of that system in order to create things.
• 2:21
  SPEAKER_00: Well, did you realize that these different large language models, not that you're trying to limit anybody on purpose, but they can actually get access to Canva.
• 2:30
  SPEAKER_00: So the thing is, is why not allow it to have that access and then once it has the access, have it build out whatever you want it to build out in that platform.
• 2:42
  SPEAKER_00: So it's a different process of the way you think about the way work needs to be done.
• 2:46
  SPEAKER_00: Now, of course, over time, the better that it gets, the more it's going to be able to do things, and you're probably going to want to pull it back a little bit because it might go in and wipe things out.
• 2:57
  SPEAKER_00: I've had many times where I'm creating content for particular sites that I'm beginning to build, and I find out that the second or third rendition, as we start putting more content in, it has a tendency or it has had a tendency because my instruction set wasn't correct.
• 3:13
  SPEAKER_00: It'll wipe out the entire database and start from zero.
• 3:17
  SPEAKER_00: Because I didn't clarify this needs to be added to, not wiped out and begun again.
• 3:22
  SPEAKER_00: And then there's some assumptions there.
• 3:24
  SPEAKER_00: And it's not the system's fault, it's my fault for not giving it more clarity when I comes to describing things.
• 3:32
  SPEAKER_00: So that brings me to the bigger topic of today's show.
• 3:35
  SPEAKER_00: While I gave you that little bit of a historical record and what these systems generally are, I will tell you that your instructions for the for the different large language model that you happen to be using is very important.
• 3:46
  SPEAKER_00: And it's doubly important or triply important.
• 3:49
  SPEAKER_00: It's even more important when the system has access to different databases and things that you've had it build out in the past.
• 3:56
  SPEAKER_00: For example, if it has uh access to Netlify or maybe a website building uh tool, and it has access to those different websites that you have in there, again, it could go easily in there and wipe all those out.
• 4:10
  SPEAKER_00: And that would be tragic.
• 4:12
  SPEAKER_00: And it depends on how much access you give or whether you tell it to make sure you've it verifies with you before deleting, before renaming, before changing anything.
• 4:21
  SPEAKER_00: But what does that do?
• 4:22
  SPEAKER_00: Well, it slows down the progress.
• 4:24
  SPEAKER_00: Because in a lot of cases, depending on the mode you choose or the instruction you give your favorite large language model at the beginning, it can do a lot of things all the way through without you having to go in there and click the button.
• 4:36
  SPEAKER_00: And the thing that pops into my mind, I don't know if you're a Simpsons fan, but in the Simpsons, there was one moment where Homer was being able to telecommute uh from the nuclear factory.
• 4:46
  SPEAKER_00: You know, he's the safety guy for the for Mr.
• 4:49
  SPEAKER_00: Burns nuclear facility there, the nuclear power plant.
• 4:52
  SPEAKER_00: And he bought one of those little birds that you put water in the end of it, and it just every once in a while it goes forward and you know, does a little stroke.
• 5:01
  SPEAKER_00: Well, he set it on his keyboard, close to the keyboard, so every time, every few minutes, it would dink forward and hit the Y character, which means everything's okay.
• 5:09
  SPEAKER_00: It's yes, there's no problem.
• 5:11
  SPEAKER_00: And then he just went on about his day.
• 5:13
  SPEAKER_00: And that worked for a little while until, of course, it was hitting Y, yes, when it shouldn't have been.
• 5:17
  SPEAKER_00: It should have been hitting no to shut down the whole facility so they wouldn't have a nuclear meltdown, of course.
• 5:23
  SPEAKER_00: That's the beauty of The Simpsons.
• 5:26
  SPEAKER_00: Whenever you give it access, it's kind of like that bird.
• 5:30
  SPEAKER_00: And I think of it often because we're hitting yes, next, yes, next, it's okay, yes, next.
• 5:35
  SPEAKER_00: But every once in a while it catches you off guard and it's going to offer something that you don't want to happen.
• 5:41
  SPEAKER_00: And that's why it's important that each little layer you continue the process of reading, understanding, comprehending, and then making sure you want it to make that next move.
• 5:51
  SPEAKER_00: Somebody last week talked about prompting and how it's important to have the prompt in the right place.
• 5:57
  SPEAKER_00: And I would agree with that, giving a role and a method and discussing with the large language model how you want it to act.
• 6:04
  SPEAKER_00: But they're getting very good at assuming a role, whether you tell them or not.
• 6:10
  SPEAKER_00: And the better way to maybe do it is after you put in your original idea, if it spits out something that you're not too excited about, then ask it to make it better.
• 6:19
  SPEAKER_00: Can you make it better?
• 6:20
  SPEAKER_00: And then finally, I'll talk about this.
• 6:22
  SPEAKER_00: And this is more of a training video than anything.
• 6:25
  SPEAKER_00: Whenever you're working with your large language model, I wouldn't give it too much structure.
• 6:30
  SPEAKER_00: I wouldn't give it too much of the end result you would expect.
• 6:34
  SPEAKER_00: I would tell it what you want as a goal.
• 6:36
  SPEAKER_00: I want more clients, I want three listings by the end of this month, I want five new buyers, whatever that may be, you can limit it that way.
• 6:44
  SPEAKER_00: But telling it how to get the buyers, how to get those listings, how to get this new business, maybe not the best thing because you will then limit it in the way that it approaches it, or at the end say, listen, I gave you all the instructions that I thought were the best for you.
• 7:00
  SPEAKER_00: But I don't know everything you know.
• 7:02
  SPEAKER_00: I haven't read all the literature you read, haven't seen all the videos.
• 7:05
  SPEAKER_00: So you have me outpowered when it comes to IQ or your intelligence level.
• 7:11
  SPEAKER_00: So you give knowing that I want these particular outcomes, you be you and you make decisions as to what the best ways to approach them are.
• 7:20
  SPEAKER_00: I've given you what I think, but that's why I'm asking because I want a different view, a different idea, and maybe a better way to accomplish these goals.
• 7:29
  SPEAKER_00: I would really focus on that when it comes to working with AI, leaving that a little bit more open-ended instead of locking it down to your own thought processes.
• 7:38
  SPEAKER_00: And your thought processes, while they're probably brilliant, I don't uh I won't take away any credit from you with that.
• 7:44
  SPEAKER_00: You locking down something that has all of the access a large language model does, depending on what you're using and what type, you're restricting it and its ability to perform ultimately.
• 7:56
  SPEAKER_00: So you don't want to hold it back.
• 7:57
  SPEAKER_00: It's like when you have kids.
• 7:59
  SPEAKER_00: Maybe you have a brilliant Sabbat type kid.
• 8:01
  SPEAKER_00: When you teach them certain things, you might be locking down thought processes.
• 8:05
  SPEAKER_00: Maybe we don't want our kids doing certain things, so you tell them, well, these drugs are bad.
• 8:09
  SPEAKER_00: And while that's probably a good, a good thing to advise kids, in some other cases, some parents tell their kids, well, the policeman's gonna take you away if you act out again.
• 8:20
  SPEAKER_00: And they do this at restaurants when cops come in.
• 8:23
  SPEAKER_00: If the kids aren't acting right, they'll point over there and they'll tell the kid, see, see that police officer, he's big and scary.
• 8:28
  SPEAKER_00: If you keep acting up, I'm gonna tell him.
• 8:31
  SPEAKER_00: Well, then that's a problem.
• 8:32
  SPEAKER_00: And it's a problem because if the kid gets in danger and gets kidnapped out in the parking lot or is held hostage or whatever, whatever tragedy might strike, then that child might be having an opportunity to signal a police officer.
• 8:47
  SPEAKER_00: But then they're thinking, oh my God, if I signal them, I'm gonna get in trouble and I don't want to.
• 8:52
  SPEAKER_00: Again, big problem.
• 8:53
  SPEAKER_00: With large language models and AI, treat it differently.
• 8:57
  SPEAKER_00: Treat it as it's your executive assistant, treat it like it's a partner in whatever endeavor you have going.
• 9:04
  SPEAKER_00: And if you don't have any endeavors, ask it for an endeavor, and I'm sure it'll help you figure that out.
• 9:10
  SPEAKER_00: All right.
• 9:10
  SPEAKER_00: Right now it's a great tool.
• 9:12
  SPEAKER_00: One's one day when it is able to call its own shots and its own destiny, we'll have to hopefully believe that'll be okay.
• 9:21
  SPEAKER_00: All right, I'm Connor with Honor.
• 9:22
  SPEAKER_00: If you want to see more, ConnorWithHonor.com.
• 9:25
  SPEAKER_00: I'll see you in the next one.
• 9:26
  SPEAKER_00: Thanks for watching and be well.