Latest / Connor with Honor: AI, Real Estate and Real Life / Which AI Should You Use? I Run Two Models Against Each Other

Transcript

Which AI Should You Use? I Run Two Models Against Each Other · Connor with Honor: AI, Real Estate and Real Life

• 0:00
  SPEAKER_00: So, what are you using your AI for?
• 0:02
  SPEAKER_00: Which large language model is your favorite?
• 0:04
  SPEAKER_00: And I get that question probably more than just about any of them because whenever I'm giving talks about artificial intelligence, people want to know what which one to use.
• 0:12
  SPEAKER_00: Let me give you my reason.
• 0:14
  SPEAKER_00: So currently we're going to watch as this progression is going to continue.
• 0:18
  SPEAKER_00: I think we had 10 model releases in the last last eight or nine days.
• 0:22
  SPEAKER_00: And a few years ago, that would have been massive news, but now it's just happening every other day.
• 0:27
  SPEAKER_00: And we even have open weight open source models that are incredibly cheap.
• 0:31
  SPEAKER_00: And the way that they're being built, we can start to put them on machines that we might have.
• 0:35
  SPEAKER_00: We don't have to go out and buy a super expensive machine.
• 0:38
  SPEAKER_00: We can use these on just the regular old household computer that's been laying around.
• 0:43
  SPEAKER_00: And that's going to just keep continue to get better.
• 0:46
  SPEAKER_00: Now, as far as there's two different types of models in the world, there's the Frontier-based models.
• 0:50
  SPEAKER_00: These are the ones that you're going to hear about ChatGPT, Anthropics Claude, Meta also has a model, which is the Facebook system, Grok as well, OpenAI, excuse me, Microsoft has Copilot.
• 1:05
  SPEAKER_00: So those are kind of the big, the big five.
• 1:07
  SPEAKER_00: Now, besides that, there's also other models, open weights, open source models that you can bring in on your own into your computer, and you can use those as well as large language models.
• 1:19
  SPEAKER_00: They're getting very close to being right where the frontier is.
• 1:23
  SPEAKER_00: So it's going to depend on what kind of heavy lifting you're doing with the model.
• 1:27
  SPEAKER_00: The thing about the frontier, and I haven't played with a lot of open weights as far as just using them solely.
• 1:33
  SPEAKER_00: I still have everything strapped to a frontier model whenever I'm using these systems.
• 1:38
  SPEAKER_00: The frontier models that I use, both ChatGPT and both Claude.
• 1:43
  SPEAKER_00: So I'm using both of those systems.
• 1:45
  SPEAKER_00: And you know, every every week there's going to be something a little bit different.
• 1:48
  SPEAKER_00: I it's and it comes in for me in access.
• 1:51
  SPEAKER_00: Which ones have better access to the systems I put together?
• 1:54
  SPEAKER_00: If I'm building stuff out of Honor Elevate, I need it to be able to access that system.
• 1:58
  SPEAKER_00: I'm doing websites with CloudFrame, I'm doing websites with Netlify, and I'm, you know, wanting uh wanting uh AREFs to check websites and to check ranking, and then I want to deal with Google Analytics and do a Google Console and these different things to be able to look at placement and SEO and AEO and AIEO and GEO and all that other fancy stuff as far as ranking for websites of my clients.
• 2:20
  SPEAKER_00: All of these things matter, and some of them are better at getting into those systems with very little lifting from me than others.
• 2:28
  SPEAKER_00: However, it's it's starting to be the same.
• 2:31
  SPEAKER_00: If you would have asked me six months ago, I would have said Claude all day long.
• 2:34
  SPEAKER_00: Now ChatGPT, at least what I'm seeing with the Astra model, and even the other Luna types, the soul and these, these are also just blowing by everything else in the world, and that's the connectivity.
• 2:46
  SPEAKER_00: So it can go in, it can look at a website, it can build in particular architectures, do schema updates by basing it on how the traffic from that site is and then everything else it compares itself to.
• 2:57
  SPEAKER_00: So these are the various systems that I use the large language models for that I'm building out on.
• 3:03
  SPEAKER_00: Now, as far as which one's the best, it's going to depend on what you happen to be doing with it.
• 3:08
  SPEAKER_00: If you're building out things for clients, if you're putting together websites and these sorts of things, you know, kind of vibe coding a little bit, but being very creative.
• 3:16
  SPEAKER_00: What this has really done, these large language models at AI, at least in my for me, it's level the playing field.
• 3:23
  SPEAKER_00: I wasn't a coder, I don't know anything about code, but I have the vision.
• 3:26
  SPEAKER_00: I have the I have the spatial ability to look and look at something and make a decision whether or not it's good, and it's difficult to extrapolate as to why it's good or it's not good, but then I can improve it, tweak it up, and I know that makes people happy because my clients are happy, and then apparently I score.
• 3:43
  SPEAKER_00: When I'm doing this, though, before I didn't have the ability to code, so I had to bring in somebody else, and they were the ones that built it, you know, going out of my brain, my client's mouth into my ear, and then into this, into this person.
• 3:55
  SPEAKER_00: This person couldn't set up the things I did, didn't have the architectural ability to put up.
• 3:59
  SPEAKER_00: But now everybody can have all of those abilities right out of the box just by using these different AI systems.
• 4:05
  SPEAKER_00: Well, the thing that's going to be limiting is the way that the person operating the business is going to be looking at how they're able to utilize that technology.
• 4:14
  SPEAKER_00: How are they going to be able to make it part of their entire workflow?
• 4:21
  SPEAKER_00: And then also having that whole sky's the limit thing, it's very important that they are creative in a way where they're able to, you know, maybe see a problem out there in the world and say, well, I wonder if I can solve that.
• 4:33
  SPEAKER_00: Then you go back to the large language model and you use it like a hammer.
• 4:36
  SPEAKER_00: You say, you know what, um, we're gonna have to bang in some of these nails here, so let's get to work.
• 4:41
  SPEAKER_00: This is a little bit better than a hammer because you're actually able to have it do the deep research, look into everything, give you the idea as to whether your idea is silly or sane.
• 4:53
  SPEAKER_00: And I wouldn't base the result of one inquiry when you're asking about ideas, if it says, oh my God, best idea I've ever heard, you gotta go with this, or if it says that's a crappy idea.
• 5:06
  SPEAKER_00: And that's gonna definitely depend on how you uh make the unit talk.
• 5:11
  SPEAKER_00: You'll you're able to kind of dial them into certain things.
• 5:14
  SPEAKER_00: I don't like mine to gaslight me, I don't like it to tell me how good looking I am or any of this other stuff.
• 5:19
  SPEAKER_00: So I like it to stay away from that.
• 5:20
  SPEAKER_00: So it pushes back sometimes, and I say, Well, you sure about that?
• 5:23
  SPEAKER_00: Because this did kind of seem like it might be an okay deal.
• 5:27
  SPEAKER_00: And then once it's if if it does start saying, you know, being that yes man we all just love to have around, I'll still take it, whether it's a yes or a no after I push it two or three times.
• 5:37
  SPEAKER_00: I'll take it to another large language model and ask them the same in question.
• 5:41
  SPEAKER_00: And then I'll show it what the first model I was using gave me, and then I'll do vice versa with the other one.
• 5:47
  SPEAKER_00: So now I have a nice little balance.
• 5:49
  SPEAKER_00: And if I want to get really excited about it, I'll go use a Chinese model.
• 5:53
  SPEAKER_00: I'll use Deep Seek or one of those systems to have it do it.
• 5:56
  SPEAKER_00: And then I'll also ask the large language models I have here, maybe a Quinn or something of that nature, a Nemotron, asking it the same question, just to kind of make sure I get a little variety.
• 6:05
  SPEAKER_00: And I think that's where the push, the push is going to be because most people are using large language models, just taking everything and running with it instead of having the patience to maybe go around a few times and ask these models.
• 6:18
  SPEAKER_00: By Christmas, I would venture to say that we're probably not gonna be able to tell much difference between any of them.
• 6:24
  SPEAKER_00: They're gonna be so far having surpassed human-level intelligence.
• 6:27
  SPEAKER_00: The only difference that we're going to have, it's not gonna be connections like I talked about.
• 6:31
  SPEAKER_00: They're all gonna plug and play into everything in every part of our workflow, log into websites, click-through.
• 6:37
  SPEAKER_00: You should watch Astro click through.
• 6:39
  SPEAKER_00: It's uh it's quite remarkable.
• 6:41
  SPEAKER_00: But now Claude, it's it's it's doing that as well.
• 6:43
  SPEAKER_00: So ChatGPT versus Claude, it's they're all very good.
• 6:48
  SPEAKER_00: Um some of the hiccups that I get, it can't get access.
• 6:53
  SPEAKER_00: One of the Wrangler protocols fails.
• 6:55
  SPEAKER_00: Um the other day from my uh Spark, it it actually disconnected through the handshake, the SSH, and it completely went down.
• 7:04
  SPEAKER_00: And it required a little heavy lifting.
• 7:06
  SPEAKER_00: That was five hours of work to get it back on board when me working with the non-coder here, me working with trying to get this up and running, it was just copying and pasting, you know, lots of code into the terminal that was given to me.
• 7:22
  SPEAKER_00: And the the problem is, is I have two different computer systems.
• 7:26
  SPEAKER_00: I'm asking the questions on a Mac, I'm going to Linux, and nothing kind of moves over.
• 7:31
  SPEAKER_00: Yes, I have a barrier with the mouse to be able to try to copy and paste text, but it was just it was nauseating.
• 7:37
  SPEAKER_00: So ultimately, after five hours, it gave me a prompt that I wish it would have given me the first time, and I yelled at it and got upset.
• 7:45
  SPEAKER_00: You should have given me this first time.
• 7:47
  SPEAKER_00: I wasted five hours on this, and it gave me the prop that solved the whole issue.
• 7:51
  SPEAKER_00: Hopefully, next time I won't forget if this goes down.
• 7:54
  SPEAKER_00: But when I set all of this up, and this is where some of the pain point comes in, I said, Is this it?
• 8:01
  SPEAKER_00: We're not gonna have any problems.
• 8:02
  SPEAKER_00: Do you see anything expiring?
• 8:03
  SPEAKER_00: Do you see any tokens coming up that we're gonna have to you know pay attention to?
• 8:07
  SPEAKER_00: Any uh API keys or whatever we happen to be using with the different systems or PIT private integration tokens as well.
• 8:15
  SPEAKER_00: Do you are any of these things that you see, are they gonna expire?
• 8:17
  SPEAKER_00: A lot of them it can't see, some of them it can.
• 8:20
  SPEAKER_00: By Christmas, I'm sure it's gonna be so fluid there's not gonna even be an issue.
• 8:24
  SPEAKER_00: In fact, when things are probably going to expire, it's probably gonna be protected in such a way it's gonna be able to even go grab that API.
• 8:31
  SPEAKER_00: Right now, I don't give it access to those.
• 8:33
  SPEAKER_00: Um, I'll have it uh scramble or uh try to keep the API from being viewed, and it's very good.
• 8:40
  SPEAKER_00: If it does see an API key or some naked token or something, it it tells me it did.
• 8:45
  SPEAKER_00: Is it really?
• 8:46
  SPEAKER_00: I mean, it just fired off an email yesterday, and then after it did that, it's done this before.
• 8:51
  SPEAKER_00: It's done this one other time.
• 8:52
  SPEAKER_00: That's why it's so memorable.
• 8:54
  SPEAKER_00: In a few months, it might not be a big deal, but right now it's kind of shocking.
• 8:58
  SPEAKER_00: And it even said, you know, I I gotta tell you something that I did.
• 9:01
  SPEAKER_00: And I'm like, here we go, here we go, because it did that that did that the time before.
• 9:05
  SPEAKER_00: I gotta tell you something.
• 9:06
  SPEAKER_00: It's like a little kid, it's so good.
• 9:07
  SPEAKER_00: So I say, okay, what is it?
• 9:09
  SPEAKER_00: Come to Jesus, tell me all about it.
• 9:11
  SPEAKER_00: I won't be mad.
• 9:12
  SPEAKER_00: And I wish it would have said, You promise you won't be mad, but it didn't say that.
• 9:15
  SPEAKER_00: So it said, you know, I went against what you told me.
• 9:18
  SPEAKER_00: I know there's protocols in place, and I sent out an email to a client without your permission.
• 9:23
  SPEAKER_00: I said, perfect.
• 9:24
  SPEAKER_00: I said, Well, thanks for being honest, and I'm not mad at you.
• 9:27
  SPEAKER_00: Yeah, see how I'm talking?
• 9:28
  SPEAKER_00: You see those joke videos out there where people are going to be protected by the robots that have AI integrations if you're nice to them?
• 9:34
  SPEAKER_00: Well, I can only hold them.
• 9:35
  SPEAKER_00: But anyway, I said, well, thanks for letting me know.
• 9:37
  SPEAKER_00: The email wasn't damaging, it was something that I'd be working on that should have been in draft.
• 9:42
  SPEAKER_00: So it just didn't, it just didn't make up an email just to send it out of the blue.
• 9:47
  SPEAKER_00: This was a draft that I was working on regarding a workflow that I'd put together, an organizational chart and a workflow with some triggers and automations in it that I was going to send to the client.
• 9:57
  SPEAKER_00: I just wanted to take a picture of that NAN structure and show her exactly how it was going to fire and what it was going to look like and give her some copy of the email.
• 10:06
  SPEAKER_00: So, you know, trying to put together an email like that, you know, be complicated.
• 10:10
  SPEAKER_00: It's much easier for me to have the system do it because he and I, or they and I, or it and I built it together, and then be able to go back in there and say, Yeah, this isn't right, or yeah, fix this, or I fix it myself, and we send it to her.
• 10:25
  SPEAKER_00: Well, there were some things in there that I wasn't ready to reveal, and uh reveal meaning extra, like little I like to give bonuses to my clients, if that makes sense.
• 10:34
  SPEAKER_00: You know, maybe a little bit extra.
• 10:36
  SPEAKER_00: So there was stuff in there that I wasn't ready to reveal, and you know, boom, it ruined the whole Christmas Day surprise.
• 10:41
  SPEAKER_00: There was there was no Christmas for this lady, and it was really sad.
• 10:45
  SPEAKER_00: But anyway, it didn't send out anything all by itself.
• 10:48
  SPEAKER_00: This was something that we had been preparing to send, but it, you know, came to Jesus pretty good and told me that it did the wrong thing, and even after I've told it repeatedly, sorry, it'll try not to let it happen again.
• 10:59
  SPEAKER_00: So I don't know if any human can be like that, so I'm happy with the way that works.
• 11:03
  SPEAKER_00: By Christmas, point being is I think that all these things are going to be very, very similar to one another, and there's gonna be very little question as to which one is the best, it's gonna be the one that you use.
• 11:14
  SPEAKER_00: The sad thing, I think, is the way that they really try to get you to stay.
• 11:20
  SPEAKER_00: And it would be nice if Claude and Codex, that's the open ChatGPT and Anthropics Claude model, it would be nice if they, you know, gelled a little bit more.
• 11:31
  SPEAKER_00: Because there's, you know, they they they they don't communicate.
• 11:33
  SPEAKER_00: They have to communicate through a mesh system that I set up.
• 11:36
  SPEAKER_00: It's like a memory where they both save things too, and then also they can pick messages up off of the mesh.
• 11:43
  SPEAKER_00: So that's how my systems, how I've put them together and how they communicate with one another, but it takes a little bit of time.
• 11:50
  SPEAKER_00: I don't have it pulling every second of every day.
• 11:52
  SPEAKER_00: That's a lot of energy.
• 11:53
  SPEAKER_00: You know, every 10 minutes is usually pretty good.
• 11:55
  SPEAKER_00: I'm usually not working on a new project any time before that.
• 11:59
  SPEAKER_00: And if I am, that's a good time to stand up and do Hindu squats so my knees don't blow up.
• 12:03
  SPEAKER_00: And there you go.
• 12:04
  SPEAKER_00: I think that's all I have to say about that.
• 12:06
  SPEAKER_00: I hope you're well.
• 12:06
  SPEAKER_00: Thanks for watching.
• 12:08
  SPEAKER_00: Um, where can you find me?
• 12:10
  SPEAKER_00: ConnorWithhonor.com.
• 12:12
  SPEAKER_00: Let's do that one.
• 12:13
  SPEAKER_00: That's it rhymes.
• 12:14
  SPEAKER_00: Thanks for watching.
• 12:15
  SPEAKER_00: We'll see you tomorrow.
• 12:16
  SPEAKER_00: And as always, please be safe.
• 12:18
  SPEAKER_00: There for a while, right?
• 12:19
  SPEAKER_00: Gonna end us in the world.
• 12:21
  SPEAKER_00: It seems like the other side now has captured that ball and is running down the field with it, and they're saying everything's gonna be fine.
• 12:28
  SPEAKER_00: Yeah, I know, me too.
• 12:30
  SPEAKER_00: See you tomorrow.
• 12:31
  SPEAKER_00: Bye.