The Immersive Lens Podcast

Paul Engin | Dave Ghidiu | Jeff Kidd

Episode 14: OpenAI

Cover art for podcast episode

 

In this episode, Paul and Dave launch their first-ever multi-part series focused on "Frontier Models" - the heavy hitters of the AI world like OpenAI, Google, and Anthropic. The discussion begins with a fascinating look at Paul’s hands-on demo of the Meta AR glasses, which feature live captioning and "bone conduction" audio, setting the stage for how AI is escaping the screen and entering our physical world. They also share a "vibe coding" success story where Dave used Claude to build a functional medical monitoring app for a doctor friend in under an hour, illustrating the power of subject-matter expertise paired with AI.

The core of the episode evaluates the OpenAI ecosystem, covering everything from the standard ChatGPT interface to specialized tools like Codex for programming and the Atlas web browser. While they acknowledge OpenAI's "first-to-market" advantage, the hosts provide a surprisingly critical perspective, ultimately labeling the current suite as "mid-tier" compared to emerging competitors. The final verdict suggests that while OpenAI is a solid entry point, the "switching cost" for users to move to other platforms is lowering, and the race for AI dominance is far from over.




Key Topics

AR Glasses are Maturing Rapidly: Paul’s experience with the Meta AR glasses highlights a shift toward seamless, non-intrusive tech; featuring "fist and thumb" cursor navigation and localized audio that only the wearer can hear, these devices are moving toward replacing the smartphone.

The Rise of the "Subject Matter Vibe Coder": Dave’s ability to build a complex EKG training app without deep coding knowledge proves that AI is shifting the value from syntax mastery to domain expertise, allowing professionals to "vibe code" solutions in minutes rather than months.

OpenAI’s Ecosystem is Broad but Fragmented: The episode breaks down the difference between standard ChatGPT, the Atlas browser for integrated research, and Codex for multi-file programming, noting that while powerful, the experience varies significantly between web and desktop versions.

The Ethical Divide in Frontier AI: The hosts discuss Anthropic's recent refusal to remove guardrails for the Pentagon regarding mass surveillance and autonomous weapons, contrasting this with OpenAI’s more aggressive pursuit of government contracts and the potential risks of AI hallucinations in high-stakes environments.





Transcript

Click here to view transcript
Jeff Kidd: Let's go. I am rolling on both.

Paul Engin: Alright. One day. One day I'll get you.

Dave Ghidiu: You already did it several, several of them.

Paul Engin: Dave, move the mic a little closer.

Paul Engin: Welcome to the Immersive Lens, a podcast exploring the technologies reshaping how we live, work, and learn.

Dave Ghidiu: From AI and virtual reality to creative media and design, we're diving into the tools and ideas shaping our connected world.

Paul Engin: My name is Paul Engin. Join us as we uncover the people and ideas driving the next wave of interactive experiences.

Dave Ghidiu: And I'm Dave Ghidiu. This is The Immersive Lens.

Paul Engin: So Dave, I had a productive weekend. I got to play with some things and one of the things that I... I talked with an old colleague and she was creating a video for her grandmother who's turning 80.

Dave Ghidiu: Okay. Okay.

Paul Engin: And I've done something like this in the past. You would ask people to shoot a little video and then they would send it to you.

Dave Ghidiu: Is it like a memory or a happy birthday greeting? Okay.

Paul Engin: And then you would put it all together. Well, apparently they've automated this so much that now it's just a web app that you literally just go to. It's called VidDay and you just have a... you register whatever, and then it sends an invite, like a regular evite. And then everyone can just go... you can leave a video message saying, "Hey, I'm doing this for my grandmother, this and that." And then can you leave a message? And they can use the webcam...

Dave Ghidiu: Oh, and then it stitches it together?

Paul Engin: And it stitches it automatically. So they don't have to do anything. And it's just these little... She showed me the interface and it was literally just like little thumbnails and you can shift them around so you can put the different orders.

Dave Ghidiu: Oh my gosh.

Paul Engin: And then... I was like, oh my gosh. I just remember doing that all manually. Like people were sending me videos or text messages and then I would have to put it together.

Dave Ghidiu: Yeah, so what does this do? I mean, it must do something. So like, your webcam obviously the resolutions are different and the volume is different. So is it doing some normalization on the back end, right?

Paul Engin: I think it's doing, yeah, and it's... and you can add some slight, subtle effects and like... it's... it just got me thinking. We talk about low barrier, and you can't get much lower barrier than that. These are people that are like, you know, I mean they're her mom's friends who are also in their 70s and 80s. So like... it's literally just clicking a link, watching a video of her saying, "Hey, can you do this?" And then they click it and...

Dave Ghidiu: And you could do that from a mobile phone too, I bet, right?

Paul Engin: Oh yeah. Yeah.

Dave Ghidiu: Oh my gosh.

Paul Engin: Yeah. So I thought that that was really interesting.

Paul Engin: And then on Friday I had an opportunity to demo the Meta Glass Display.

Dave Ghidiu: Tell me what that is.

Paul Engin: So that is Meta's glasses that are augmented reality, where you get a little small display in the right eye.

Dave Ghidiu: Okay, so it's not over the whole lens, it's like a corner of the upper right hand corner or something of that lens?

Paul Engin: Yep. Okay. And then you put a wristband on for tracking of your hand gestures.

Dave Ghidiu: You were talking about that when we were talking about like Selfridges when you went to England. Okay, yeah, yeah, yeah.

Paul Engin: Yes. So, but those were just the Meta glasses with the AI that I could just talk to. This is actually like with a display.

Dave Ghidiu: Oh, okay.

Paul Engin: So, I was... first of all, this process is crazy, but I know that it's an early adopter thing. But they only have it... you're not allowed to purchase it unless you do a demo first. So I had to find a place that had a demo, and it's nowhere near us. So I had to go do the demo. And then they did the demo on a black pair of glasses, and I said, "Oh, cool. Okay, I'll take this," thinking, okay, I'm leaving here with glasses. And he was like, "Oh no, these black ones aren't even shipping for several months." And I'm like, "Okay." He goes, "But we can start shipping these beige ones now." And I'm like, "I don't really care about the look. I'm looking at the functionalities." But, um... I was like, "Okay, so can I grab those now?" And they're like, "Oh, no, no. You're going to get it..." so I'm not even going to get it for like several weeks.

Dave Ghidiu: Oh man.

Paul Engin: But the experience was amazing. I put them on, and I did the quick thing which was the live captioning. So I can see closed captioning as people are talking. And it's focus. So if I had multiple people talking, if I turn my head and I focus on you, it's captioning on your focus.

Dave Ghidiu: Come on. I kind of want those just to navigate life.

Paul Engin: Right? Well, and again, this is only going to get better. I mean I just keep saying like this is... But the crazy thing was I was able to make my hand in a fist like this and then use my thumb as a cursor. So I just go forward with my thumb, I skim it to the left or right...

Dave Ghidiu: So your thumb is basically like a mouse kind of?

Paul Engin: Yeah. And I was navigating apps, and like I could go to a music app, and I could just, oh, and then I would pinch my fingers to select it.

Dave Ghidiu: Okay, much like on the Apple Watch. Okay.

Paul Engin: Yep. And then it would just launch and I'd be listening to music. But you can't listen to it unless you're like right behind my ear, like if you're back here so...

Dave Ghidiu: Is it bone conduction?

Paul Engin: Yeah, it's something with the... I love bone conduction. I was like, wait, you can't hear what I'm hearing? And he was like, no, I can't hear it. And I was like, this is crazy. And then, it allowed you to do translation obviously if you needed to do it.

Dave Ghidiu: Sure. He only knew English, the guy that was giving me the demo. So you could if I was speaking for instance Spanish to you, it would the glasses would translate.

Paul Engin: Oh my gosh. Yeah.

Dave Ghidiu: Now you've been very like secretive and clandestine about this whole thing, so it was like at a remote location that's not close to here, and...

Paul Engin: Well, I could tell you what it was if you want.

Dave Ghidiu: I'm more curious like is the price palatable?

Paul Engin: No. Not right now. Not right now. Okay. But, you know what it'll be is I think it'll be good that you and I can like look at it, see the usability of it, see the feasibility of it. I think there's a lot to it. I think there's a future in this 100 percent. I won't need to get my... when I have these glasses on, I won't need my phone.

Dave Ghidiu: I am betting dollars to donuts that the reason you have to have like an audition or an interview before you can buy them is because the Apple headset, a lot of people returned those because maybe they didn't understand all the features. So like you kind of went through this training, this onboarding, and so now you see the value and you're like, oh okay, I get what I'm buying, so you're less likely to return them.

Paul Engin: Yeah, I could see that. I mean, you could turn the display off too and it could just be audio too. So I could be like, you know, "Hey Meta, what's the blah blah blah blah." And it would tell me and it could talk, and I could have a dialogue with it. Um, just like I did like the one uh in London when I used that.

Dave Ghidiu: Yeah, and so I have bone conduction headphones and that's all I use anymore. I don't use buds or anything like that. Because I like hearing the outside world, but also um, there's nothing in my ear. So it sounds like I would just use those in lieu of those at the very least, right?

Paul Engin: Right, exactly. So I think that there's so much potential once we get it and I get to play with more of the apps. The other thing that the version they had didn't have was they had a handwriting so you could, if I wanted to write something, I could just pull out my left hand and then with my right hand, I can literally just pretend like I'm writing on it.

Dave Ghidiu: Oh that's awesome. And one of your predictions in an early episode already came true when you were like we're not going to need these special bracelets, like it will just be a watch band. And I just read an article that said like that's what they're doing now. So...

Paul Engin: That was crazy.

Dave Ghidiu: So Big Tech come recruit Paul, he's got the vision.

Paul Engin: Well it's just a natural and you know with the Apple Watches they're interchangeable so you can just slide in and out and I'm like well rather than wearing this, I'm surprised Meta doesn't... well...

Dave Ghidiu: They will.

Paul Engin: You know, give it something for Apple or someone will.

Dave Ghidiu: No, Meta will just have a Meta watch.

Paul Engin: Yeah, that's true. That's true.

Dave Ghidiu: Um, so I have a few things that just happened and I kind of just want to talk about.

Paul Engin: Yeah.

Dave Ghidiu: One is because we're hot off the heels of the Open Devin episode.

Paul Engin: Yeah.

Dave Ghidiu: And so Open Devin is a just reminder is this fully autonomous like AI bot that you can unleash on your computer and will have unfettered access to everything you do. And the head of Meta's AI alignment, the director of Meta's AI alignment...

Paul Engin: Yeah.

Dave Ghidiu: Had it on her computer. And despite repeated commands to stop, it erased every email in her mailbox. And went through the trash and like ultra deleted it and then apologized like, "Yep, you definitely told me not to do that, and I did it. I'm super sorry." So that's just a cautionary tale.

Paul Engin: Oh my gosh.

Dave Ghidiu: Well, you know what, we're going to come back to that in a minute for something else, but I just want to say kind of like how I spent my weekend. I was... one of my buddies, Aaron, he works, he's a doctor in one of the local emergency departments. And his kids were in this wrestling match. He's like, "Hey Dave, do you want to come watch my kids wrestle?" I was like, "Yeah, sure." So we were there and we were only there for maybe an hour, and he was like, "You know what, I have this idea for an app. Can I bounce it off you?" It was basically he wanted to train people how to use like an EKG. And you know there's like the you know "do do, do do," the things that you see on like every... right.

Paul Engin: Yeah, beep, yeah, yep, yep.

Dave Ghidiu: Yep, that's the one. And it has like the EKG and then it has like the I don't know the pulse ox and like the blood pressure and all that stuff. So he's like, "I have a monitor, and this is what I want it to look like. Can we make a web app so that I can go to the web app and get a special link where I can control it, and then someone else gets just the display. So they just see the spiky graph thing and the vitals." I was like, "Yep." So in under an hour we made that happen. Uh, with and I know I talk about Vibe Coding, but I've upped my game and I'm now using Claude Code, which sits inside an IDE. And an IDE is a super nerdy thing you see on TV whenever people are like programming and historically it was like, that's just us clacking our keyboards and like writing code. But now there's kind of two panes, the code, and then adjacent to that is Claude, which is so like it lives right in that IDE. And I can be like, "I want an EKG app that does this." And the prompt was actually quite lengthy.

Paul Engin: Yeah.

Dave Ghidiu: And first the first delivery, boom. It worked out of the box. And then we spent some time tweaking it. And like this is something I never would have done because A, I never would have dreamed it up, and B, even if I made this, I have no idea like, is this realistic? Is this working? Like, I don't know what any of this is. So because I was working with a subject matter expert, Aaron was like, "You know what, like..." and he used all these terms, so I will not do the conversation justice. But he was like, "We need to have like the deflection of the PQRST or whatever that is..." And I was like, "I don't know what you're saying," but like I typed it into Claude and Claude knew, and we were able to together kind of navigate this. He's like, "I don't know the web technologies," so I was like, "Well, it's PHP because we need server-side." He's like, "See, that's the stuff I don't have." So this is just another reaffirming notion that, or reaffirming thought and experience that what we're doing with Vibe Coding here at the college, because we have a Vibe Coding course, like this is where we need to be because this is like, you can't do it yourself as a Vibe Coder, and you can't do it yourself as someone who knows nothing about web technologies. But together, boy oh boy, it's really, really cool.

Paul Engin: And I just saw the app and it is really, really cool. The simple fact that you can...

Dave Ghidiu: Yeah, tell them about the phone.

Paul Engin: Yeah, you... Dave took the phone, he had a QR code on the screen, he took a picture of it, and then he basically had like the instructor station so he could determine the different levels of heart... and I don't know any of the terms, so I'm just going to say heart rate for lack of my knowledge. Uh, and so it was increasing, increasing, and then there's a certain time when you can shock it depending on the heart rate. So he's controlling basically what the patient is.

Dave Ghidiu: The vitals, yeah, yeah.

Paul Engin: The vitals. And I was like this is brilliant. How many places could use this? And so you could do it live. So he could say, uh, "they're going about the flat line, I'm going to flat line them, and then see if they know where to shock or when to shock."

Dave Ghidiu: Yeah, and so this is my last episode because I'm going to break out and do a medtech company using AI apps, so bye everyone, it's been fun.

Paul Engin: Awesome, awesome.

Dave Ghidiu: So it was really fun. And I do have two errata. I was bingeing our episodes um, over the weekend.

Paul Engin: Uh-huh.

Dave Ghidiu: And one, I accidentally said Joseph Conrad when I meant to say Joseph Campbell talking about the Hero's Journey.

Paul Engin: Oh, okay, okay.

Dave Ghidiu: And then also I was talking about how a lot of a notebook I think has the... I call them disjunctions, like the "uh" and the "um" and the podcasts, and it's really called disfluency. My brother, who I have a twin brother who's a software engineer, he was he's like, "Dave, you're an idiot, it's a disfluency."

Paul Engin: Man, calling you out like that.

Dave Ghidiu: I know. So we have had some fan mail, and we've had my brother.

Paul Engin: Telling you that you're great.

Dave Ghidiu: Yes, telling me that I'm great. All right, so what are we going to talk about today?

Paul Engin: Uh, so today we're going to talk about um, well actually, do we want to talk about what happened with Anthropic CEO?

Dave Ghidiu: Yeah. So give us the high altitude version of... So just to anchor this, this is March 2nd, so we're talking about events around like February 26th to March 1st.

Paul Engin: Right, right. So um, Claude right now, Anthropic has an AI LLM that's called Claude, and they already are implemented into the government. So they created a custom version of Claude specifically for the government.

Dave Ghidiu: For classified missions too, right? Like the Maduro Venezuela thing was... yep.

Paul Engin: Everything, yeah. They're doing from administration to I mean they had tons of different applications. Um, but this past week they gave them like a three-day deadline or something to adhere to two things that they weren't agreeing on. So they basically agreed to 98 to 99 percent... he says, um...

Dave Ghidiu: You're talking about Dario Amodei.

Paul Engin: Yeah, Dario, the CEO, told an interview, I think it was with CBS, um, that they agreed to 98 to 99 percent of all the use cases that the Pentagon wanted. But they didn't agree to two. And the two were mass surveillance of American citizens and full autonomous weapons.

Dave Ghidiu: So that's like drones that will go around and like kill targets without human intervention.

Paul Engin: Right, it just makes the assessment itself and it does it. Um, and it was really interesting because what the Pentagon wanted is for them to remove these two guard rails. And Dario basically said that he feels that the surveillance on American citizens kind of breaks the democratic values that we're trying to represent. And which is it's good, it's an ideology that I think is important, but I think the other thing he mentioned and this goes to the autonomous weapons is that the technology is not there yet. And he alludes to kind of like, think about the hallucinations and things that happen just when we're doing things that aren't like...

Dave Ghidiu: Yeah, like making up citations?

Paul Engin: Yeah, right? Um, so the systems aren't there yet to say that there shouldn't be human intervention because it could hallucinate something while it's actively, if it's an autonomous drone attacking a school, thinking that it is, um, it sees a lot of people moving around in and out and maybe it thinks that like, there's all these factors that you know, and I'm simplifying it, but you just never know what it will do yet, so it's not there yet. And that's where his point was, in the future it might be, and they might be open to it, but it's not ready right now.

Dave Ghidiu: Yeah, and I think back to all the huge errors where it lied about citations or it can't even do math very easily, and I'm like, oh really, we're going to allow it to fly autonomously into like a war zone and make those types of decisions?

Paul Engin: Right. I mean, I think about even when we were doing ChatGPT and I was like, "Hey can you do this?" And they're like, "No, we can't do this, we can't do this," and I say, "But you just announced you can do it." "Oh yes, you're right, we did just announce we can do it, but I can't do it." I'm like, okay.

Dave Ghidiu: And it makes me think of like when ChatGPT 5.1 or 5.2 came out, they were talking about em dashes because that's like the latest rage. And Sam Altman was like, em dashes will no longer appear. And even if you told ChatGPT like that week, like no em dashes, it'd be like, okay, and then the next thing you know you're getting em dashes. I'm like, if you can't control the em dash, like, you should not be flying a plane.

Paul Engin: Right, right. So, um, you know, regardless of what you think about it, I thought it was a great stance for an organization that really has these strong beliefs and...

Dave Ghidiu: Yeah, their PR people must be just so ecstatic because this is hot off the heels of the Super Bowl where the Claude, the Anthropic Claude ads were poking a lot of fun at OpenAI because of like ads coming in...

Paul Engin: Well it wasn't directly at OpenAI. It was, you know, just a coincidence.

Dave Ghidiu: All the other AI companies that have ads in them. And so like now this, this is the... and OpenAI was, I'm sorry, Claude was or Anthropic was founded by people who kind of splintered off of OpenAI because they didn't believe in the ethics that were happening then. So this is like five years ago. Um, so this is just yeah another testament to the ethos of Claude and Anthropic and everyone there. So good for them, good on them.

Paul Engin: Yeah, yeah, we'll see what happens. I'm sure at some point we'll talk about...

Dave Ghidiu: And I think that leads into today because OpenAI quickly kind of like sees the opportunity like, "Oh okay government, we'll be your LLM of choice."

Paul Engin: Yes. As of right now, yeah.

Dave Ghidiu: So I figured what we'll do is we'll kick off a series of our top frontier models and kind of...

Paul Engin: A series?

Dave Ghidiu: Yeah. This is our first series.

Paul Engin: Jeff, do we have any theme music for this series?

Jeff Kidd: Let's see what I get running on it.

Dave Ghidiu: All right. So we do, we have theme music for this series.

Paul Engin: So um...

Dave Ghidiu: The Peaky Blinders... We're doing the watch party? For the yeah. So this is season one episode two of Peaky Blinders.

Paul Engin: Yes. So the series that or the frontier model we were going to talk about today is going to be OpenAI.

Dave Ghidiu: And the frontier models basically, this is going to be a bit of an abstraction but it's when we talk about LLMs or like chatbots or like services that people pay for, the frontier models would be Gemini, ChatGPT, Claude, Perplexity, Meta. So we'll be looking at some of those.

Paul Engin: Yep, yep, absolutely.

Dave Ghidiu: And so frontier basically means like these are bleeding edge, cutting edge, high compute time, they're doing all the work.

Paul Engin: Yes. And they're the most dominant ones right now that we can think of. And there's a lot of smaller LLMs out there, but these are the ones that kind of are building and branching and they're moving away just from a chat bot, so those are the things that, you know, we'll talk about today.

Dave Ghidiu: Ah, okay, okay, yeah.

Paul Engin: Um, so I guess the first thing that everyone's most familiar with is obviously the chat the chat itself, right, the ChatGPT. And that is just a standard prompt and it's both online and it's an app that you can download. And for everything we talk about today, for the most part, there is a just web browser based login or you can download an app on your device both on mobile and on your computer and run it locally. And it's interesting because you have a different take on this.

Dave Ghidiu: I do, I prefer web apps. So I never install software on my computers. Um, it's a vector of attack, it's just one more thing to maintain, it makes things go slower, and the biggest thing is because you don't always have the latest and greatest version, whereas if you go to the website, they're always serving you the latest version. So that is just my nerdy nerd, nerd nerd position.

Paul Engin: And I ran into that this weekend when we started talking about that and uh you know I might be pivoting to that. Although it's interesting because they always say download the app so you can do more.

Dave Ghidiu: Yes, but that apparently evidently is not always true. The other thing is like with Chrome and presumably other browsers, you can turn a website into an app so that it becomes a web app that acts as an app.

Paul Engin: Alright, that's like Inception.

Dave Ghidiu: Yeah, say that again. Um, you can take a webpage and turn it into a web app.

Paul Engin: Uh-huh.

Dave Ghidiu: So then you get the little icon down in the sys tray, and you get native functionality, so keyboard commands that would normally like Ctrl P for instance in Chrome prints, but if you turn it into a web app, if Ctrl P has a different function, then it's overridden, it's overriding the print function and will do whatever that app wants.

Paul Engin: Gotcha, gotcha. Okay.

Dave Ghidiu: But that's a conversation for another day.

Paul Engin: So yeah, so obviously there's the standard prompt and that prompt can do a lot of things. And I think most people are familiar with it. Um, and so I was thinking what we could do is we could talk about some of the other derivatives of that GPT chat.

Dave Ghidiu: So when we say GPT chat, we're talking about like ask a question, get responses, generate images, that kind of stuff.

Paul Engin: Yeah. So those are like table stakes across like all these frontier models.

Dave Ghidiu: Yeah, they all have them. Yeah.

Paul Engin: They are. They are. And it's funny, I think as there it's being commoditized is like, all right, well I don't know the different... you could do Gemini, you could do, you know, it's a weird...

Dave Ghidiu: And for a while, there really was differentiation, but now all the image generators, for the most part, I mean Google actually just dropped Nano Banana 2, but generally speaking, they will all get you where you need to go. Um, and so the differentiation now is in kind of these auxiliary services and the other things that these models can do.

Paul Engin: Yes. And so with all these models too, you have the option of obviously doing a long thinking, a shorter thinking. Um, there's premium purchases. Um, one of the things that is I think is kind of nice about it is there is a temporary chat inside of the... I don't know if it's on the web, is it on the web app too?

Dave Ghidiu: Oh yeah.

Paul Engin: Um, the application, if you download ChatGPT, if you at the bottom there's a little pull down, it's like right now it says 5.2, but if you click it, if you turn on temporary chat, that will create a something that won't store in history, they won't supposedly it says won't be used for training models, um, so they keep it around for 30 days and then it gets deleted. So it was an interesting little addition that I mean it's been there for a little bit, but I thought that was interesting for people who don't want to share content. I still don't think you should put confidential stuff in there, but...

Dave Ghidiu: And Gemini and Claude also both have this like ghost mode or temporary chat.

Paul Engin: Where they won't be trained on it either, right? Because I think there's also a distinction of tracking history and...

Dave Ghidiu: Yep, and actually Claude, by def, well not by default anymore, but you can opt out to have all your conversations not trained on.

Paul Engin: Oh.

Dave Ghidiu: Yeah. We'll get to that in another part of our series.

Paul Engin: Um, and it also gives you the option to search on the web or just within its own LLM model. Um, so you have some additional triggers that you could do with it. Um, the one thing that was interesting to me is that you know we talk about having dialogue and this is one of the things that I you do it regularly but I wanted to have a dialogue with ChatGPT, so I hit the mic and I was like "da da da da," but it didn't talk back to me.

Dave Ghidiu: Yeah, it won't talk back. You well, there might be a live mode in there, like Gemini and Claude both have a live mode. I imagine ChatGPT does, you might need to have it on the phone. I don't know if the computer has it.

Paul Engin: Okay, it is on the phone. The phone definitely has it and you could talk to it and you can have a dialogue back and forth. Um, but this downloaded app does not right now.

Dave Ghidiu: Yeah. And what's your take on this?

Paul Engin: I prefer talking to LLMs. Uh, typing is for chumps, man. It just goes a lot faster. So if you can finagle it, and there are apps out there like Whisper Flow, I think we talked about. Um, and since we talked about it, it has been available for Android, so I started using it, and I really like it. It doesn't solve that problem that you're talking about where it talks back to you. Uh, but I mean in my car or when I'm walking the dogs, I'll have my headphones on, I can have conversations with Gemini, and we can brainstorm things and talk about things or I can just kind of like pick a topic and go down a rabbit hole. So...

Paul Engin: Yeah, I'm surprised though. I thought that like cause Copilot has it. And that's Microsoft's, and again it's maybe another LLM we can look at. But with Microsoft I'm able to just click, have a dialogue, it can look at my screen, it could do more than... I was really surprised that this in the downloaded app I wouldn't have...

Dave Ghidiu: Yeah, so it has so it can access all that information.

Paul Engin: Yeah. Um, but uh you were exploring AI agents, right? Creating agent builder?

Dave Ghidiu: Yes. The agent builder. So we're going to talk I think a little bit about like Atlas, Codex, and like part of the suite of like features. So the agent builder and an agent is kind of a squishy term in AI. Microsoft for instance says like chatbots are agents, which most people disagree with that. But agents typically, and Emily Laird, who's someone I follow in the space is like, well it has to have like seven or more steps that it can do in a row. But if it can act on your behalf and accomplish multiple tasks to do a thing, so for instance, if you task an agent with scheduling a haircut and then a dinner reservation, so it has to like look at your calendar, it has to see if you're available, it has to like figure out who you like getting your haircut by, so that is kind of the dream of being an agent. And I did look at the OpenAI agential workflow, the builder workflow, it is not something I would recommend to the faint of heart. It's very it's somewhat complicated. Uh, it does cost money. Not a lot. Like I mean you pay per token usage. Um, and you can and like I put five dollars down to just kind of like play with it. But you need some technical know-how to kind of navigate it. And I showed you the left hand screen, you know where it usually has like all the different tools, and you just scroll and scroll and scroll and scroll. There's a lot going on there. So I don't think that that's for people like you and me. Yeah. I think that's for like if our college hired someone to make agential workflows within the college, they'd have to really understand kind of how all the systems plug in and APIs and hooks and all that stuff.

Paul Engin: Yeah, and it looked like it was a node-based, so if you picture these little blocks, and you basically can connect one block to another block to another block. So...

Dave Ghidiu: And it has the branching, like true or false, like is this person, should this person go to the next stage? So yeah, it's good, but it's I can't even imagine what I'd use it for in my personal life.

Paul Engin: Yeah, and I think that's the big thing is understanding use cases for these. I know like they build, you know, you hear these things where, "Oh, I'm going to build my marketing team." And it's like supposedly it's supposed to do all of these things that you know somebody would do in marketing, and I could see that to some extent, but they would have to have control over a lot of aspects like it's not just conceptual. It's like, okay well open up Adobe Express or Canva and then create a blog, and then do a write up, and then do like it's not the email campaign, it's not just like in a prompt it's got to use other applications to do.

Dave Ghidiu: Yeah. And it still makes me like, I really don't want to surrender my credentials to anyone and I mean I will talk about the Meta AI, you know it just deleted every email in her mailbox. So like I'm not so sure I'm ready for this agential AI, but I think with supervision and with time it's going to get a lot better.

Paul Engin: Yes, yes. And I think it will and it'll simplify. Like even the little the video thing, think how simple it got, you know, and so I think that that is something that we should keep exploring but you know figure out how it can be integrated. I think it'll be more meaningful.

Dave Ghidiu: Yeah, and I think that we're actually seeing kind of an erosion of the line between agents and say regular LLM usage, like Claude Code now kind of acts like an agent because it does multiple things. You give it a general task and it will do all these small tasks to do it. Uh, or Claude Co-work. So we're starting to see the line is blurred where general AI usage is becoming like micro-agential.

Paul Engin: Yeah. And we can kind of talk, why don't we go into Codex?

Dave Ghidiu: Oh yeah, what is Codex?

Paul Engin: So Codex is the and it's interesting because it's really the coding part of ChatGPT, so it focuses on coding. And it's an API that can be plugged into um Visual Studio Code or you use it on GitHub.

Dave Ghidiu: Claude code. Oh yeah, yeah, yeah. GitHub.

Paul Engin: GitHub. Um, and so these are just applications, these are IDEs that you can like typically without the AI that's what you would code in. So we would open up a coding window and we would just code and you could test. And now it's integrated into it, so now you Dave alluded to this when he did his coding is you have your code window, you have well you have your directory, code window, and then now you have your chat. And Codex is living in that chat. So...

Dave Ghidiu: And so you just type in the chat and then the code window gets filled with the code that the chat...

Paul Engin: Yep. And then you could do testing on it and all of that stuff.

Dave Ghidiu: Yeah, the debugging is really really nice. In making test cases, things like that. So Coda, and I think Codex is actually really slick, right? It is a good experience.

Paul Engin: Yes, I've I had a really good experience. I did a web app that did analysis on the market and it did different more risky stocks versus safer stocks and you can do an assessment over one or two years. And it's all live data. And again this is where I had to get an API from a stock exchange that I could get the real-time data and I had to plug that in. I mean it told me what to do, but there was still that like if you're not familiar with like any of these terms we're talking right now...

Dave Ghidiu: Yeah, like API seems scary, like, "Oh my gosh, that's like nothing," and it's not hard. Like you go, you put your email and then you get the API, which is kind of like a key so they know. It's a big number.

Paul Engin: It's a big number. You copy and paste it in. And it's associated with your account with them, so...

Dave Ghidiu: So it's nice because it walks you through those steps that even if you don't know what an API is or even where to get one or where to put it once you get it, it kind of walks you through all that stuff.

Paul Engin: Yep, absolutely. And then it's like little things that people might not know, the site has to be live, it has to be functioning. So what like if you don't know, it'll tell you, "Oh, you have to do an Apache run and it creates a local server, so then it can run the application." But if you didn't know that, it tells you about it.

Dave Ghidiu: Yeah, and that's nice. The one thing about that is so someone who encounters that, they're like, "type this in to get an Apache server," you can say like, "I don't know what an Apache server is before I type this in. Can you just explain this to me?" And it will walk you through it.

Paul Engin: Yes, yes. And then um, I think Codex also added something called Codex Automation, um, where it can basically generate more automated things that can run on a regular basis. So if you were to do like a, you wanted to schedule something that checks your folders every now and then or run a code check or you know, it can update code on the fly weekly if that's what you want or branch it on a regular basis. You can run that automation, and schedule that to run weekly, daily or however...

Dave Ghidiu: So doing a lot of stuff that a developer would do.

Paul Engin: Right.

Dave Ghidiu: Okay. And when I was at the NYAIC conference last week, there was a presentation from ChatGPT, from OpenAI about Codex, and the gentleman presenting was like, "just so you know, AI did a lot of the code base for Codex." So like even though Codex is software, AI programmed that software. And it's the first time we've had like that level of success.

Paul Engin: Yeah, that was really interesting 'cause it's basically like it can automate bug fixes and refactoring, so it can on a regular basis you can say like all right every other day I want you to go through the code that's created and so you can have an automated I don't want to call it agent but a Kodak automation that checks code that you use that you're creating on another like it's not...

Dave Ghidiu: Yeah, it's like another employee. It's like, "okay Dave, here's what I want you to do every Monday morning, do this and go through it and make sure that things are running smoothly, refactor the code and make sure like run these new tests."

Paul Engin: Yes, yes. So it was really interesting. I think that this is a great tool for anybody who's interested in like you know, what you did, you had an idea and you wanted to quickly develop it and you were able to just develop it using this tool. And Chat GPT does it too, but this is more geared toward...

Dave Ghidiu: Yeah, the difference is usually when I program in Chat GPT or like Claude or Gemini, if you're just doing one page, that's fine, you can copy and paste that, but if you're doing 10 pages, you definitely want to be using Codex, if you have a whole...

Paul Engin: Because you're linking files.

Dave Ghidiu: You're linking files and it can look across all those files at the same time.

Paul Engin: Yes. Um, so the other thing I wanted to talk about was uh we touched on it last time, I downloaded Atlas.

Dave Ghidiu: Atlas. And remind me what Atlas was.

Paul Engin: So this is OpenAI's web browser.

Dave Ghidiu: So it's a browser.

Paul Engin: It's a web browser. And I was under the impression, and this was interesting if you go back to the episode, I thought that it would be able to log you into systems and do things for you automatically, like automated.

Dave Ghidiu: Yeah, that's the dream, isn't it?

Paul Engin: That is the dream, but it was not, that it doesn't do that. So it's not.

Dave Ghidiu: Wah wah wah.

Paul Engin: It's not allowed to log into systems like that and automate stuff. But so I was trying to like and I don't know if I explained this right, but I was trying to get the difference between okay well then what's the difference between let's say Atlas and just a GPT prompt, right, the Chat GPT prompt.

Dave Ghidiu: Yeah, if it's not a browser, what is it? What's it doing?

Paul Engin: Right. So it is a browser. And the difference is rather than having the like let's say Chrome open and then having Chat GPT open, and then I find something on the webpage and I copy and I paste that into the chat and I ask it questions, or I find something and I put the URL into the chat, it's all integrated. So when I'm on a site, I have a tab on the left, I can just quickly say, you know, "hey look through the site, tell me the summary of whatever, you know, whatever I want for that site," and it'll scan the whole site, and I don't need to leave the browser.

Dave Ghidiu: So it just makes it easier, you're not copying, you're not pasting, you're not losing things, it's all it can it has the intelligence for whatever web pages are open or whatever tabs you have open.

Paul Engin: Correct. And that was the other use cases. Technically I could be on Target looking at like TVs, I could go to Best Buy look at TVs, Walmart look at TVs, and all three tabs could be open, and I'm just trying to find the best price. Well rather than me doing that, I could just simply in the right say it recognizes all the tabs that are open and I can say, "Please find me an 85 inch OLED TV."

Dave Ghidiu: Are you really looking for an 85 inch TV?

Paul Engin: No, I'm not.

Dave Ghidiu: Okay. Does anyone want to get that for me? Yeah, we should get one for the pod. Yes. We should get two. Right here somewhere. I don't know if we could open the door.

Paul Engin: But so it could do that comparison and it could give you the summary based on the tabs that are open. So it's just, it makes it a little bit more fluid as far as working with it.

Dave Ghidiu: Yeah, so it's like seamless integration of AI browsing experience. So it's not going to click buttons for you necessarily, but it will make kind of parsing and assimilating all that information a little bit easier for you.

Paul Engin: Yes, and it's all integrated, so I'm not leaving the browser and going to another app and then going back to the browser.

Dave Ghidiu: I could see the value in that, that seems that seems nice.

Paul Engin: So. Yeah. So I felt like I was like okay that's that's interesting it's it's not as overbearing as I thought as far as like oh my gosh it's going to take over this or do that you like...

Dave Ghidiu: So it's very subtle.

Paul Engin: Yeah. Okay. Very subtle to start out.

Dave Ghidiu: Low, no barrier.

Paul Engin: Yeah. Um, so Atlas is their web browser which makes browsing easier. Codex is their coding platform that makes coding easier. Yep. And then I figured we I was exploring GPTs.

Dave Ghidiu: And that's like the custom GPTs, like custom chatbots.

Paul Engin: Yeah, so you so first of all this was an experience in its own. First you need a Plus to create your own.

Dave Ghidiu: And that's like 20 bucks a month, right?

Paul Engin: Yeah. You need at least a Plus account. And then ironically this is where the the download app versus online app comes in. I had the app downloaded and I said okay I want to create a GPT. And I said yep just click on explore and click create. And I went there and there was no create button. I'm like, "there's no create button." And it said, "oh you don't have the the Plus version." I go, "nope I have the Plus version." And it said, "oh well you got to go online to create." And I was like, "So you need that web app, baby."

Dave Ghidiu: Well it's hard to to maintain the code base for a web app, for the PC version, for the Android version, for the iOS version, for the macOS version. So that's another reason why the web apps are better because you only have one codebase.

Paul Engin: Yes. So yeah so I finally got there, and there's a bunch of GPTs that already exist. Um, so you can...

Dave Ghidiu: So these are like custom bots that people have made?

Paul Engin: Yeah, that that, and there's a bunch of different ones from like a video AI, scholar GPT, image generator...

Dave Ghidiu: And I made one a few weeks ago that was basically and it's not hard to make one, like I don't want to seem when I say the word program it's not hard to program, you just like give it instructions in plain English. But I made one where I could upload an image and it would spit out the alt text because I wanted to make that for like our push for accessibility at the college.

Paul Engin: Right.

Dave Ghidiu: It was easy to do and then it was easy to deploy.

Paul Engin: Right. And so the way you should think, for people who are, it's something that is a task that you might do multiple times, rather than having to give the same instruction over and over and over to the Chat GPT, you could create a custom GPT and tell it these are the instructions you want to happen. And then every time you open that you can now just say like for you Dave, I assume now you could just upload your image, you don't have to give any prompt, and then it'll just give you the alt text.

Dave Ghidiu: Automatically. There's no because you already set up what you want it to do.

Paul Engin: And it's also great for reports. So say the department of planning is going out and they have to do inspections all the time. You know they can just, you can program the bot to ask it like the eight questions with the responses that and then they plug in the responses and then it format like you upload a sample and it will format the report for you. So I think using it for repetitive tasks is a great way to use it. And if you're listening to this and you're like nah that sounds a little comp just try making your own bot. And so right now we're talking about Chat GPT and the custom GPTs, but that also exists in Gemini they're called Gems, and we'll talk about that a little bit next week, but...

Paul Engin: And so I created one for our podcast. Um, and so now what I can do is when we have a topic that's coming up in the you know like if we're going to do Claude next week, I'll be able to put that topic in and it will kind of give me a give us a breakdown based on the information that I put in there and give us suggestions and now that's something that we could just keep reusing. And it's cool because it's something that you can make public or that you can share, so like I shared a link with you so technically you could do the same thing and get the same information.

Dave Ghidiu: Yeah. And one of the nice things a difference between this and a chat is, and I might be wrong on this, but I think I'm right for at least for Chat GPT, that it doesn't really retain the memory, it's meant to be a tool that does the same thing, but not be swayed and persuaded by previous chats that you've had with it.

Paul Engin: Right, it still remembers the chats, but it doesn't take that into account unless you copy and move it into, it's called its knowledge base.

Dave Ghidiu: Okay.

Paul Engin: So you can move it into its knowledge base and then once it's there then that's what it pulls from. So...

Dave Ghidiu: So if I wrote a ChatGPT, or if I made a ChatGPT that was like I'm going to give you a topic and I want you to make a haiku, then say the first time I use it I do say write a haiku about yoga. The second time I do I'm like write a haiku about Legos. I'm not going to get a haiku about like Legos with a yoga flare.

Paul Engin: Correct. Right. Unless you copy and paste it and keep adding to it. And that's I think really interesting because you can really build out these GPTs because as you find things, you can add it to the knowledge base, as you get artifacts you can upload that knowledge base, you know, and it can be different types of files, word documents or whatever artifacts, images. And it will... when you get the custom chat, it'll be specific to that. So like ours analyzes the transcripts, and looks for retention leaks. It rewrites the titles for maybe more increased downloads, so it gives us suggestions for titles. And then it can talk about how it can find short clips that we could cut up to make social media posts. So it was enlightening to me to see how automated we could make some of this stuff.

Dave Ghidiu: Why do we even do this? Let's just have AI do it.

Paul Engin: I know. I know we can get ElevenLabs to do the audio.

Dave Ghidiu: I do like that. So like we you and I and Jeff, we could go through the videos and like find that the cool things that we want to put in the YouTube Shorts or whatever. But like this can do it, and do it faster and probably better than we could.

Paul Engin: Yeah 'cause it's interesting because it can do an analysis on what is being talked about and then compare it to what is trending or what is relevant.

Dave Ghidiu: Oh my gosh, that's cool.

Paul Engin: And it can like and especially if you're like I want to target this specific demographic, it can even break out sections for those demographics.

Dave Ghidiu: I didn't even think about that because so I'm not really big on social media so I don't have Instagram, I don't have TikTok or anything like that. Although I do have YouTube Shorts. But I have no idea what's trending. So you're telling me that we could upload say this episode and say hey, look at what's trending right now. If Paul and Dave talk about anything that's trending, then this is the the slice we want to put in the social media.

Paul Engin: Yeah, it will give us the slice based on the transcript.

Dave Ghidiu: And this is this is 20 a month you need to pay 20 a month to get this feature?

Paul Engin: Yep. That's worth right.

Dave Ghidiu: If you're to create your own GPT. Yeah. The I mean and you can create as many as you want, right?

Paul Engin: Yep. Well I assume so. I've only created one.

Dave Ghidiu: Okay. So you can create at least one. I know that you can do at least four or five. So this is I mean if you're out there looking to be like well I have 20 bucks a month that I could spend on some type of marketing budget, like this is... yeah. Yeah, I think that this would be worth exploring because or you know even if it's not marketing, even if it's like a task I was thinking about doing our advising, you know we have we could just upload our course our course sheets, we can and courses, and technically you could put information in there and it would, you know, spit back obviously you have to be careful. This is where it's like you don't how much of a person's information do you want in there. Yeah. Because I don't know if these are trained, that's what I'm not sure about is if they're trained on your custom GPTs. Yeah. We'll have to dig in that and find out. Although in maybe that's a use of like agential AI, like that might be worth the investment to like if we're looking for like a course advisor or something like that. Yeah. Oh yeah, you're right. Yeah. That's that's interesting. Yeah.

Dave Ghidiu: Uh before we land this plane, is there anything else you want to talk to us about with OpenAI?

Paul Engin: Um, I think the only thing we didn't talk about was Sora, but we talked about it previously. It's just their video generation model, and it's still available. You can still generate videos. Um, I haven't played too much more with it since I've been exploring other things, but that's still in place. And you know, I have my favorite with Flow because I feel like I have the Google one because I have more control over it. And but you have the experience too. If I was sitting in front of Flow, I'm like, well I don't know what I'm going to do with it. Yeah. So I think that this is good for like maybe B-roll or something, I just don't know if you can do consistent characters yet with this or any of the stuff that you could do with Flow. So and again, I need to explore it again because maybe there's some more things that they've added that since I've used it.

Dave Ghidiu: Yeah, and I'd like to rank like give a ranking for OpenAI just in general based on like my experience with their chat, with Codex, it's mid. I think it's fine. I think the strength of OpenAI is that they were first to market and that was kind of you know their moat is that no one else can touch us because we have all the users. And quite frankly when Gemini first their first attempt was called Bard and remember it was terrible, people were like I'm not going to switch, I'm not going to jump ship. Yeah. And so like the switching cost now is kind of high, which is why I think a lot of people still use ChatGPT even though I don't believe it's the best frontier model out there. But, you can export your chat conversations from OpenAI and Google is rolling out import chat conversations. So if you are looking to switch to say Google or Claude, and I'm sure this will be kind of like a standard in the industry, the switching cost just got a lot cheaper.

Paul Engin: Yeah, yeah, that's a great point because if you're going to go from one to the other, you have a history with one.

Dave Ghidiu: Yeah, so if you've been using it since November of 22, you have like three and a half four years of chats that you can now bring with you and then that becomes part of whatever frontier model you're using in the future.

Paul Engin: Oh god, great point. Yeah, that's a really good point. I mean it's the same thing with migrating data from one to another. Yeah. That's interesting because something that if you just started getting into it you don't think about, but because it's there it would be great to be able to export it. Yeah.

Dave Ghidiu: So I gave Chat GPT a rating of mid. What's your rating?

Paul Engin: Um...

Dave Ghidiu: And you can parse it down into different sections like Codex and Sora and stuff.

Paul Engin: Yeah, so I would say Atlas was interesting. And you could do that in any platform, but I thought that was interesting. I was I went to Marques Brownlee the YouTuber, I went to his channel and I did a while I was in Atlas and I said analyze this entire page what patterns or what patterns do you see in titles, thumbnails, and upload cadence, summarize the channel's content strategy. And it gave me a really great breakdown of everything. And I was I was really impressed that I didn't have to leave the browser.

Dave Ghidiu: Yeah, so I'm looking at the notes that you took because you took screenshots and it was like, "Series branding: strong recurring series. Dope tech explained. First impressions reviews." And MKBHD is a fantastic technology channel to follow. He's awesome.

Paul Engin: And the... I met him.

Dave Ghidiu: You did not.

Paul Engin: It was right after COVID, my son and I were flying to one of his hockey events and he was in the airport, he was masked, but...

Dave Ghidiu: You could...

Paul Engin: Yeah, we took a picture with him. Well we were on a it was a weird trip, we were on a plane with another group of YouTubers that I wasn't familiar with, my son was, oh my gosh. And then all of a sudden we got off the plane with them and they were they were pretty rowdy. And then they saw Marques and they all ran over to Marques and he's like, "oh we follow you all..." And I was like, "Oh my gosh Jack, that's Marques." My son's name is Jack. But we so we ended up I was like, come on, you know he's embarrassed of me, but I'm like come on we're going to go get a photo with him so.

Dave Ghidiu: Sweet.

Paul Engin: Yeah, so that was that was cool. But yeah he's amazing, his channel's great, he does great reviews, really really good platform. But anyway the point with Atlas was that it gave me a pretty good summary, pretty seamless. So you can do market research then with like Atlas, like if you own a company and you're like, hey here's my competitors, you go to them, you're like what's their competitive advantage, what's their branding. Okay. Cool.

Paul Engin: Yep, what could we do to yep. And again you could do the same thing in a regular chat, but this is just integrated so seamless so. I think that I'm going to rate all of this as mid as well. I'm not... like the Codex was really good, but I've used Claude too and it felt better. Yeah. Okay.

Dave Ghidiu: And so that's a good teaser for staying with the series because we still have Gemini and Claude at least to go through. Yeah, I think Chat GPT is fine, they were first to market, they had some great potential there, but I think that they don't excel in any arena over their other competitors.

Paul Engin: No, they did, like their video model that Sora model was amazing when it first came out, but then it just got trumped by, you know, another model, and we'll we can keep going through all the different models. But right now you know what we can do is at the end we can rank where we think they're at afterwards. But yeah, I'm going to say this is in the mid right now as well.

Dave Ghidiu: Okay. It's mid.

Paul Engin: Yes. Alright. So that's all we have time for today. And again, we'll continue the series with the next podcast and we'll have to figure out if it's going to be Claude or Gemini, what are we thinking?

Dave Ghidiu: It's going to be a surprise, folks.

Paul Engin: Alright. You gotta stay tuned. My name is Paul Engin.

Dave Ghidiu: I'm Dave Ghidiu, and if you enjoyed the conversation today, please be sure to subscribe and like, and subscribe and love, so you never miss an episode. Let's be careful out there, folks.

Paul Engin: And share it with a friend and colleague. Until next time, stay curious, stay connected, and thanks for looking through The Immersive Lens with us.

Dave Ghidiu: This episode was engineered by at least Jeff Kidd and/or Hugh Laird. I saw both of them in the booth.

Paul Engin: Yes. They were both in and out. I don't know team effort. Yeah, I don't know who's still here.

(Laughter)

Dave Ghidiu: And we'll never know. Too early, too early. Recorded at Finger Lakes Community College Podcast Studios located in beautiful Canandaigua, New York, in the heart of the Finger Lakes region, offering more than 55 degrees, certificates, microcredentials, and workforce training programs.

Paul Engin: Thank you to Public Relations and Communications, Marketing, and the FLCC AI Hub.

Dave Ghidiu: Eager to delve into your passion? Discover exciting and immersive opportunities at www.flcc.edu. Only three Ws. Three Ws are there. As part of our mission at FLCC, we are committed to making education accessible, innovative, and aligned with the needs of both students and employers.

Paul Engin: The views expressed in this podcast are those of the hosts and guests and do not necessarily reflect the official position of Finger Lakes Community College.

Dave Ghidiu: We should probably start off with that one. I think that's a pretty important one. Yeah.

Paul Engin: Music by Den from Pixabay.

Dave Ghidiu: This is The Immersive Lens.

(Music fades out)

Paul Engin: Oh. Was that Hugh or was that Jeff? It's like Statler and Waldorf back there in the booth somewhere.

Dave Ghidiu: Or Beavis and Butthead.

Paul Engin: I have a question. Are those manufactured like digitally or are those like analog?

Dave Ghidiu: It's a soundboard.

Paul Engin: Okay. Do you have a sad trombone in there? 'Cause we could we could really use a sad trombone.

Dave Ghidiu: I've heard a sad trombone.


full-width