Artificial Intelligence: Real Talk

[00:00:00] **Jethro:** All right. Welcome to whatever podcast you're listening to. I'm Jethro Jones, and today we have Aaron McKelky, who is, a former Wyoming educator who now works, for a San Francisco startup called Descript.

[00:00:14] **Jethro:** And, that is my tool of choice when it comes to editing podcasts. And, he got invited to go to OpenAI's Codex 5.5 party that happened on May 5th, and so we're gonna talk about that today, and I hope you enjoy this. And we got some deep stuff at the end of the conversation, so make sure you listen all the way to the end for that.

[00:00:39] **Jethro:** Here's my conversation with Aaron McKelky.

[00:00:42] **Jethro:** Well, I think the first place to start is tell us, Aaron, about going to the, is that appropriate to call it Mecca? Is, is OpenAI enough of a cult that you could call it a Mecca or something else? I mean, how was it going to the mothership?

[00:01:02] **Aaron:** I was gonna say the mothership is what comes to mind for me. Yeah. it was the weirdest night of my life, and equal parts just didn't know what to expect. I had nothing that I could compare it to in my previous life experience. But also it was a very open-ended thing, so they didn't really tell us what to expect.

[00:01:24] **Aaron:** But it was bunch of nerds in one room with no clear itinerary and lots of food, and everybody had 10X rate limits on their codecs. So they're not only using it and like power users, but I think intentionally a few days before the event, they 10X'd everybody's rate limits that were coming to further juice the Twitter flywheel and our, you know, use cases of codecs.

[00:01:55] **Jethro:** Yeah, so this was a literal party that was created, for people to attend, and you went all the way from Wyoming to San Francisco to OpenAI headquarters and, And so were people, like, using Codex in the moment on, like, while they were there? Instead of chatting with people, they were chatting with their AI still?

[00:02:19] **Aaron:** was impressed at the social skills that the social media people had because one of the thoughts going in was, "Okay, they're very extroverted on Twitter, most of those who got invited. They're warriors that have no shortage of opinion when they're behind computer." But how will they be in real life? And, I, I was impressed actually. Most of them were fairly extroverted and competent social skills. yeah, you're f- it's funny, a lot of the questioning, we had a few persistent questions that every little cluster of people would ask each other, and one was, "Are you running codecs right now? What, what are you, what are you using it for?" And

[00:02:58] **Jethro:** Yeah.

[00:03:02] **Aaron:** before I left." And I wasn't that hardcore 'cause my laptop was in my backpack. But mine was just that I gave it a Telegram token and I said, "When I'm away, I don't do the whole put your thumb in your laptop and keep it from closing or enter the terminal command that keeps it awake."

[00:03:20] **Aaron:** I just, I gave mine a, an inbox, so if I'm on mobile I can capture things and it's voice native Yeah. The other one was how are you gonna hit the 10X limits? 'Cause if you're on a pro plan,

[00:03:33] **Jethro:** Yeah.

[00:03:33] **Aaron:** 20X, and then they 10X'd your 20X, and it became this weird unofficial competition of can you even hit those rate limits, and if so, how do you do that?

[00:03:44] **Jethro:** Dude, I mean, the thing is, I have the $100 a month plan, and I d- I have not hit the rate limits on Codex yet. I did hit them regularly on, on Claude when I was on the $100 plan there, but not-- I just, I haven't been able to do it, and I feel like I am putting big tasks in front of it, and it's still not getting close.

[00:04:08] **Aaron:** Yeah,

[00:04:08] **Jethro:** did anybody hit it?

[00:04:11] **Aaron:** I have only heard of one person who claims that they have, and it was five projects in parallel on extra high fast, which uses double rate limits.

[00:04:23] **Jethro:** Yeah.

[00:04:23] **Aaron:** it as it was more of an attempt to just burn them than, "Oh yeah, this is what I would normally use it for."

[00:04:29] **Jethro:** Yeah. Yeah, it's interesting, and I feel like I use it pretty heavily. Like, I've made a ton of sites and done a bunch of stuff that I would think would take a lot, and it doesn't seem to be taking as much as I thought it would, to be honest.

[00:04:46] **Aaron:** Yeah, especially if you come over from a Claude plan like most people have.

[00:04:51] **Jethro:** Yeah, that makes sense. So, so I wanna hear a little bit more about this party. One of the things that I thought was the coolest, from your, X post is they had a, a photo booth, and it-- you'd take a picture, and then it would redraw that picture with different styles based on what, ChatGPT Image Maker, whatever that's called, can do.

[00:05:15] **Jethro:** So, that I thought would be a great thing. My daughter's graduating high school, and that would be a super fun thing to do at their, at that grad night party, is just take pictures and have it remake everybody. I think that'd be pretty fun.

[00:05:31] **Aaron:** Yeah, and it was cool how they set it up because on the outside of the booth you could see a scrolling view of previous people's images. And when you finished taking your pictures inside, you got a QR code, so you could actually open on your screen and see the generations in real time. You didn't have to wait until you got out and saw the physical printout. So you got to leave with it on your phone, and you could easily hit save to camera roll for each. Plus you got the printout, plus yours is now in a collage that's scrolling on the outside of the booth.

[00:06:04] **Jethro:** Yeah. I- that's such a fun novelty-type thing that, like, why wouldn't you go in there and do that? It would just be a fun thing to do. I mean, that, that kind of stuff is, is the kind of thing that, that I think makes it really personable and approachable for anybody. And it makes them think, "Oh yeah, I could, I, I could see myself participating in this," where not all AI-related things are that way, you know?

[00:06:36] **Aaron:** Yeah. Well, 'cause if you've been in a f- normal photo booth, you've already done everything required to make this one work. You stand there and it says, "Three, two, one," and it takes a picture. That's all you had to do.

[00:06:47] **Jethro:** Yeah, that's pretty simple. all right, what else? Anything else interesting or exciting or unique or whatever?

[00:06:56] **Aaron:** Yeah, I mean, I think the process of how they picked people is very... They said, "This is how we did it," but a question a lot of us inside the event was, "Why do you think you got picked?" If you weren't one of the big influencer YouTubers, which I am not, and most of the people I met were not. I think

[00:07:18] **Jethro:** Yeah, you keep saying, you keep saying you're not, and yet you sure have a lot of opportunities and things happening that make it seem like you are. So maybe you actually are, Aaron.

[00:07:30] **Aaron:** No, not, not compared to... There are people that would walk in and you would know beyond the Twitterverse that person from their YouTube channel or I follow them. There were some of those that probably got, I don't know, VIP invites that the algorithm or the model didn't. But I mean, I met a guy from Nebraska who had 400-something followers on Twitter, and you go, "Okay, from the outside, why? paid for your flight and your hotel. It was not to get good PR on Twitter," 'cause you're not big on Twitter. You're not trying to be. There had to be some element of randomness or just he's done cool things with Codex even if he doesn't have a big following or a lot of reach. So

[00:08:17] **Jethro:** Yeah,

[00:08:20] **Aaron:** picked. I won't quote people, but another thing that I heard was that there was a quota for and that when they prompted it to pick from the 8,000 or 9,000 people who applied to the event, that there was a minimum percentage that should be female, which I thought was interesting. a- and actually, some of the females in the room, they were clearly the minority. I would say, yeah, about 20%, anecdotally is what I noticed, some of the most influential. They were-- One of them interviewed Sam Altman with a, a mic and a camera and got him on her channel or whatever she did with it. So they, they performed well in person, and I can see why they did that.

[00:09:04] **Aaron:** 'Cause maybe Twi- Twitter usually does skew male when you look at your demographics.

[00:09:10] **Jethro:** for sure. That's interesting. And, you know, that's, that's fine. Whatever. I mean, I'm, I'm, I'm less curious about the gender requirement and more curious about the other things. Like, what, what even puts anyone into that category of, "Yes, we should invite them"? And, you know, it, it was rumored that the AI was the one choosing that, and maybe it was, maybe it wasn't.

[00:09:37] **Jethro:** But certainly somebody was prompting it saying, "Choose things based on this." Any other, theories about what that was?

[00:09:44] **Aaron:** Yeah. Again, like off the record, but what I was told was part of that prompt was people who make Codex approachable or are

[00:09:54] **Jethro:** Mm-hmm.

[00:09:54] **Aaron:** things, and they intentionally said and the reach should be little to no part of your basis.

[00:10:05] **Jethro:** Yeah.

[00:10:05] **Aaron:** just people sharing cool stuff that would inspire others to build with it is what they were really looking for.

[00:10:11] **Aaron:** Which think for me personally, that would make sense. I don't have huge reach or lots of followers. But if you go back to February, I've posted, especially on Twitter, quite a few things about, "Here's how I use Codex. Here's why I think it's a good value. Here's what I'm building with it." And that was,

[00:10:28] **Jethro:** Yeah.

[00:10:29] **Aaron:** not to get invited to some party that didn't exist yet.

[00:10:32] **Aaron:** It was just documenting what I do. So yeah, I think that made sense. I also-- There were international. There was a lady from Norway or Sweden. There were some rural states. I love that Wyoming, Nebraska... I guess Minnesota's not as rural, but you don't think of that with tech.

[00:10:51] **Jethro:** Yeah. Yep. Even though 3M is based there, so there is a huge tech piece there. Same with Wisconsin with Epic that does My Chart.

[00:11:01] **Aaron:** Yep. were college students that hadn't graduated yet. One Purdue student who was brilliant and pitched Sam Altman on his little app that ChatGPT should integrate. He, I think, had finals this week, the week after the event, and then graduating. There was

[00:11:20] **Jethro:** Wow.

[00:11:21] **Aaron:** student who had 10 days until high school graduation, and I asked him, "What was the response from your classmates, and teachers, and principal?"

[00:11:29] **Aaron:** And he said, "Well, most of them didn't know who Sam Altman was, but when they heard it was AI related, they kind of scoffed at it."

[00:11:36] **Jethro:** Yeah.

[00:11:36] **Aaron:** And I said, I said, "Your, your trip is gonna be way more interesting than the, the speeches at graduation and whatever

[00:11:43] **Jethro:** Oh, yeah.

[00:11:44] **Aaron:** promoted by your school. I would wanna talk to you about your trip more than hear the valedictorian at some Minnesota high school graduation."

[00:11:53] **Jethro:** Yeah, no kidding. That's pretty cool. I didn't realize that it w- it included students like that too. That's, that's very good. So, how many people were there? I, I think you said it was very small, like 100 or so?

[00:12:08] **Aaron:** Yeah. Yeah, I think it was between 100 and 200. They didn't really tell us. There were a lot of security. So just going in the front door, I was the first one at the venue, and there were two guys outside, and they basically told me, "Get out of here for a while. You're too early." but then when I came back and got to be the first one to go through security, there were probably 25 security just in the lobby alone. And I wanna say their office is, I don't know, at least 10 floors. So we got marshaled by security in groups of 10, and then they would put us in an elevator with a security guard, so we were kind of or escorted all

[00:12:52] **Jethro:** Yeah. Yep.

[00:12:54] **Aaron:** to get to the correct floor, and then walk to where the food and the drinks were, then there was security there.

[00:13:00] **Aaron:** And Sam Altman, I would guess, had probably four personal bodyguards that were hovering around him from different angles everywhere he went. So I have not been in a, in a setting like that or waited in line while people are kind of side-eyeing you thinking, "Is this guy a threat?" Or,

[00:13:20] **Jethro:** Yeah.

[00:13:21] **Aaron:** person safe to get to the front of the line?"

[00:13:24] **Jethro:** So did you get to meet Sam Altman?

[00:13:27] **Aaron:** I had to wait probably four different times, and I tried to be polite. The story of how I finally got to talk to him, there were maybe 15 minutes left in the event, and I had waited before, but he'd say like, "Please excuse me, I'm gonna go get a drink or grab some food." Which I actually took as, "I just need a mental break," and he was

[00:13:47] **Jethro:** Uh-huh.

[00:13:49] **Aaron:** 'cause people are taking selfies and pitching him on whatever. And I had noticed one of the s- one of the bodyguards had been on the same side of the room as me before. So with a few minutes left, I'm

[00:14:03] **Jethro:** Yeah. Well, and, and that

[00:14:07] **Aaron:** a selfie." And I struck up a conversation with the one bodyguard.

[00:14:11] **Aaron:** I said, "Hey, it's cool if I don't get to meet him. I was just happy I got to come. I'm glad you guys were here so we could come in, 'cause I know without security, this wouldn't have happened." And I was just making small talk with him. And when the person who was meeting him at the time finished and took their selfie, the bodyguard put his hand on my back and pushed me to the front of the line and said, "Mr.

[00:14:33] **Aaron:** Altman, this gentleman's been waiting, waiting very patiently. I think you should speak with him next." So I felt special 'cause from what I saw, I was the only one that got guided up by somebody else to talk to him. The rest were fighting in line to get a chance to, which was that idea of sometimes the most important person in the room is not the one with the most fame or money.

[00:14:55] **Aaron:** Like, I don't think I would've got to talk to Sam if I wasn't nice to his bodyguard first.

[00:14:59] **Jethro:** is a, a key piece of information for anybody who is, is working with anybody that just because that person is not the person doesn't mean that they don't have power in a situation. And, you know, I, I've learned that numerous times, from different people who, you know, you treat everyone involved with respect, and that makes things so much better for you and everyone else and the people who are there.

[00:15:31] **Jethro:** Like, that is a really important thing. So you can't, you can't ignore that idea and putting yourself in a position to help people, taking their pictures, like that just is a generous way of being. So when you actually got to go meet him, what was that experience like?

[00:15:48] **Jethro:** And did you take a selfie?

[00:15:51] **Aaron:** I had somebody else take the picture, and I felt kinda gross doing it, but then

[00:15:55] **Jethro:** Mm-hmm.

[00:15:56] **Aaron:** knew, how could I not? I had this

[00:15:59] **Jethro:** Yeah.

[00:15:59] **Aaron:** dialogue about that. But my buddy who made the merch had these wood-stained, key chains with the Codex logo, and I had

[00:16:09] **Jethro:** Mm-hmm.

[00:16:10] **Aaron:** them. giving them out to attendees.

[00:16:12] **Aaron:** I saved the, the last one I had to give away, I saved for Sam, and I just thought everybody else

[00:16:18] **Jethro:** Yeah.

[00:16:18] **Aaron:** here asking him for something or pitching him. So I just

[00:16:22] **Jethro:** Or my blessing of their

[00:16:27] **Aaron:** to combine Wyoming's culture and AI and tech, which are the two worlds that I live in, and I wanted you to have this. And I think he was a little bit taken aback, like, "Oh, you brought me something? I thought everyone else wanted my picture or my autograph." And so he laughed

[00:16:45] **Jethro:** app.

[00:16:46] **Aaron:** Exactly, and

[00:16:47] **Jethro:** Yeah.

[00:16:48] **Aaron:** and yeah. And then we talked about... He, he said he loved Wyoming and that he lives on a ranch, and I'm-- This will not shock you, Jethro, 'cause you know me pretty well.

[00:16:58] **Aaron:** But I, I pushed back on that a little bit and I said, "A ranch in Wyoming?" And he said, "No, like an hour and a half out of the city." And I said, "Yeah, it's, that's not the same, especially not the same tax bracket." And he

[00:17:09] **Jethro:** No, for sure.

[00:17:11] **Aaron:** he gave me a little playful punch in the arm, and then the only serious talk at, I guess I'd love to hear what you would say. But if you had one conversation for somebody like that, who's shaping the tool that a lot of the world is being built with, it's almost like how electricity was at the end of the 19th century or the internet was in the '90s. I mean, that's what AI is, and they're not the only company, but they're one of the two big players. My topic that I brought up with him was I said, "Hey, I come from an education background." The first couple of years, I think there was a lot of heads in the sand and no one cared. now AI is getting a bad rap, and the common response is ban it, block it, it's evil, it's unethical, it's only good for cheating. And while education does need to change, I've that before AI came out, I thought that before Google came out, I still think that. have to help change the narrative around AI more intentionally because what high school kids see on TikTok or college students think it's for, we both know is not the only way to use it, which is cheat, shortcut, substitute. And, you know, you bear some responsibility for that, such as shaping the narrative to be more, you want to learn, it gives you superpowers. But if you want to cheat, for now, it also gives you superpowers, and that's how most people are choosing to use it.

[00:18:38] **Jethro:** Yeah, that's, that's interesting. I-- Number one, I think you did a great thing by giving a gift instead of, asking for something or, asking, like,

[00:18:50] **Aaron:** Yeah. Well, and somebody who's

[00:18:50] **Jethro:** trying to get something from him. So what, what I think that does is that opens the door for a conversation rather than a, give me advice or endorse my app or, or whatever.

[00:19:03] **Jethro:** when, when I've had opportunities to meet people that I think are, playing at a, a higher level than I am, one of the things that I typically ask is, something along the lines of, "If, if you were in my shoes, what would you be focusing on? If, if you weren't where you are right now and you were somewhere else, what would you be, what would you be focusing on right now?"

[00:19:27] **Jethro:** So that's one. Another one that I often ask people is, "What was the, the thing that you learned in the last two weeks, month, whatever, that you think you'll remember for a really long time?" And I, I try to get into things that are a little bit more deep than, than just like chit-chatting type stuff. And, so that's, that's the approach that I take.

[00:19:57] **Jethro:** I don't know if that is good or bad or what, but that's kind of how, you know, in answer to your question, what would I do if I met him? Honestly, I don't even know. But I kind of think one of those things.

[00:20:10] **Aaron:** public spotlight, they've answered questions on their vision for the

[00:20:16] **Jethro:** Mm-hmm.

[00:20:18] **Aaron:** all these things working in an AI native world that they're trying to build and He-- I haven't heard him talk much about education, and I guess fundamentally that's what I see everything as is education, and there's such a big gap with

[00:20:33] **Jethro:** Yeah.

[00:20:35] **Aaron:** versus what people think you can do, or what they know how to do versus what it could actually do.

[00:20:41] **Jethro:** Yeah. Yeah. Well, and I, I think that almost everything goes back to education as well. So, you know, whether it's formal education or not formal education, it all goes back to people learning how to interact with these things and figuring out what they're for. so how, how was your, wood Codex logo, like, received by everybody?

[00:21:05] **Jethro:** Sam didn't slap you and say, "This is a violation of our trademark. You're not allowed to give these out," or anything, did he?

[00:21:12] **Aaron:** store, their merch store has been sold out of everything for months.

[00:21:16] **Jethro:** Yeah.

[00:21:16] **Aaron:** was gonna push me on that, I was gonna say, "Maybe if you had something in stock I could have bought for the trip, I would've, but I had to

[00:21:22] **Jethro:** Yeah.

[00:21:23] **Aaron:** my own, man."

[00:21:24] **Jethro:** Little,

[00:21:25] **Aaron:** eh, they-- I think most of them thought it was cool. I also...

[00:21:29] **Aaron:** In the age of tweets and the internet and AI, people feel like things are ephemeral, so that's why I thought something made out of wood that you can hold in your hand, and it just got stained two days ago, so it still smelled very strongly of

[00:21:43] **Jethro:** little tacky still? Yeah.

[00:21:45] **Aaron:** Yeah, I thought that was like a cool contrast to a bunch of AI nerds that are thinking about models and tokens and of that stuff. No. one thing that happened, I shared in this group chat, some of the OpenAI employees were in it and a lot of the Twitter people. A few days before the event, I took a picture of the key chains and I said, "Hey, when we're there, find me. I have a gift for you." And then what ended up happening is I would meet people in the lobby, 'cause I got in about an hour before it started, and I'd say, "Hey, I'm the guy who brought the key chains."

[00:22:21] **Aaron:** Well, then they would put it on their lanyard or their... Somehow display it, and the people coming in would say, "Oh, that's really cool. Where'd you get it?" And then they'd point at me and say, "That's the key chain guy. Go ask him." So

[00:22:33] **Jethro:** Yeah.

[00:22:37] **Aaron:** conversations that I could see them pointing at me, and it wasn't, "He's the smart one," or, "He's the best at Codex."

[00:22:43] **Aaron:** It was, "He's the key chain guy."

[00:22:46] **Jethro:** Yeah. Well, you know, there are worse things to be known by. And did you put, like, a little, X, @ whatever your X handle is? Are you s- The Aaron, is that what it is now?

[00:22:58] **Aaron:** That's... That is.

[00:22:59] **Jethro:** Yeah. Yeah.

[00:22:59] **Aaron:** no, but with the hat that my buddy made. Shout out Wyoming Roots, Art Ireland is his name. It's a little hard to see 'cause it's silver in most places, but The exchange I made for my keychain was, "I need your Twitter handle signed on my Codex hat

[00:23:20] **Jethro:** Oh, there

[00:23:21] **Aaron:** as a thing to put up in my office." And no one will ever see that and care besides me, but just as a

[00:23:27] **Jethro:** Yeah.

[00:23:29] **Aaron:** So

[00:23:29] **Jethro:** Yeah

[00:23:30] **Aaron:** Altman's Twitter handle autograph, I guess you would call it.

[00:23:34] **Jethro:** Mm-hmm. Look at that. Amazing.

[00:23:39] **Aaron:** Yeah. I

[00:23:40] **Jethro:** Cool.

[00:23:41] **Aaron:** things, you know?

[00:23:42] **Jethro:** Yeah, there's, there's value in tangible things, and I have, this, coaster that my friend Danny Bauer made, and it says Transformative Principal on it, which is cool, but this is on my desk every single day and probably will be for as long as I live, because it's cool, and why would I ever get rid of it?

[00:24:05] **Jethro:** So,

[00:24:06] **Aaron:** And right here in front

[00:24:08] **Jethro:** Hey.

[00:24:09] **Aaron:** recording spot, I have a sweet notepad that's for my ideas that aren't digital worthy or I want a physical copy while I'm here in my office, and that's in my top left drawer.

[00:24:24] **Jethro:** Yeah, look at that.

[00:24:25] **Aaron:** I...

[00:24:26] **Jethro:** both, we both like physical things, and I think that's, that's very worthwhile.

[00:24:31] **Aaron:** Yeah. A, a topic I do wanna discuss, 'cause everybody there asked each other the same question. There

[00:24:39] **Jethro:** Uh-huh.

[00:24:39] **Aaron:** questions across groups, but one we all asked each other was, "What do you think the point of this was? Like, why did they do this?"

[00:24:47] **Jethro:** Yeah.

[00:24:48] **Aaron:** And I, I have no answers. I have my own thoughts, but ranged from PR the lawsuit with Elon is going on, you know,

[00:24:59] **Jethro:** And

[00:25:04] **Aaron:** Another one was, are they just using us as free marketing? So instead of paying a bunch of influencers to tweet or post videos about a thing, it was cheaper to fly us

[00:25:14] **Jethro:** you're like, "Hey, I'm looking for

[00:25:19] **Aaron:** it, and it's purely like a marketing stunt for the company and the app. Is it product feedback? Because they did have quite a few employees working the floor that would walk up and say, "Hey, I work on product Codex. What's one thing you wish it could do?" Or, "What, what's your favorite use case?" And, the person in charge of the Codex plugins was there, and my company had just submitted a Codex plugin that day. So there was definitely some of that going on.

[00:25:48] **Jethro:** this."

[00:25:49] **Aaron:** Yeah. I followed up with him. I DM'd him afterwards and

[00:25:53] **Jethro:** Yeah.

[00:25:54] **Aaron:** reject it, let me know why." But, it definitely-- I don't think it was a, a marketing stunt, because why would you bring in people Nebraska, Wyoming with reach. So I didn't think it was that.

[00:26:09] **Aaron:** I also don't think PR-wise, the, the normies of the world don't pay attention to that stuff, and it was on Twitter, so it, it wasn't like they were trying to get out to Instagram and mainstream media and Facebook. So... And they're not dumb. Twitter's a weird echo chamber of tech stuff, which you know very well. So I, I genuinely think it was probably a whimsical, like, "Oh, I don't know. Let's just do it. Let's just, for fun or for science, let the model pick," which I think there was truth to. But I also think it was something like, what if you got all your biggest fans and power users in one room and just let them talk and shove drinks and hors d'oeuvres in their face and said, "Have at it." And then your own employees kinda listen in and ask questions along the way. That's, that's what I think it was primarily. It was more like a power user get-together feedback and networking session that transcends a Zoom or

[00:27:12] **Jethro:** Yeah.

[00:27:14] **Aaron:** a Discord channel. It's just not the same as, "Oh, we'll fly you in from Norway and Florida and New York."

[00:27:21] **Jethro:** Yeah, I do think there's definitely value there in that aspect of it, and it certainly is marketing, as in this came out of the marketing department's budget, I would imagine. You know? Like, that's the category that it falls in, because how else do you, how else do you justify it? Probably not product, because they're...

[00:27:45] **Jethro:** Maybe feedback could be considered that, but it's not really that either. You know? Like, that happened for sure, but that's probably not it. but definitely marketing. This is the kind of thing that you would spend marketing money on. And even if it's not, like, it, it's not the same as advertising, right?

[00:28:05] **Jethro:** But it's not exactly marketing either. And so you, you have the opportunity to do these different kind of things to get people talking about what you're doing. And like you said, this is probably a cheaper, easier, faster way to get a bunch of positive buzz about something than, like, just doing a contract with all 100, 150 of you to post something good about Codex or OpenAI or whatever.

[00:28:36] **Jethro:** Like, that's, that's cool and all, but this gives you a real experience that, you know, they can learn a lot from. From who interacted, how you interacted, what kinds of questions you asked, what kinds of- Things happened while, while you all were there, and it definitely is a valuable thing for them. And you know how difficult it is to get people who use a product to tell the company that makes the product what they think about it, even when people love it, you know?

[00:29:09] **Jethro:** It's still not that easy to get that to happen. So that sounds like the, the exact right thing for them to do to get some sort of feedback. And now that you know someone personally, you actually feel more likely to reach out to them and talk to them, and to say, "Hey, Jethro, you're working on this thing.

[00:29:28] **Jethro:** Why don't you talk to this guy who actually works at the company and see if there's something there?" And if you've built a relationship, then, it's easy to make that kind of an introduction. If you have no relationship, then it's like, yeah, email support, and that sucks. So yeah. What, what was the... Oh, go ahead.

[00:29:47] **Jethro:** Yeah.

[00:29:48] **Aaron:** I was gonna say, from the AI skeptics or the people who specifically don't trust OpenAI or Sam Altman, maybe they're like a Claude person or they say Claude is the ethical AI lab, which I'm not here to dissuade anyone of anything. A thing that I've heard was, "Oh, did they censor what you could post or when you could film or what he would answer?" And there was, you know, a joke that was made was like, "Did you kiss the ring to get in?" And it was a- a- an assumption that, like, to go there and fanboy by choice. But I can't think of a stronger way to point out that that's not the case, imagine all the people who got invited were whispering to each other like, "Why do you think they did this?" And

[00:30:36] **Jethro:** Yeah.

[00:30:40] **Aaron:** with other attendees was, "Why do you think they did this?" And it wasn't... You know, we were all skeptical of it, and part of it was I, I'm not sure why they invited me. Why do they think they invited you? But I, I also heard Altman say multiple negative things about GPT 5.5.

[00:30:57] **Aaron:** a person was going on a rant about what they didn't like, and he would say things like, "Yeah, you're f- you're right. That's fair. It is terrible at that." And so I did not get the impression that it was a authoritarian, censored setup.

[00:31:14] **Jethro:** Mm-hmm.

[00:31:15] **Aaron:** the employees I talked to were all super friendly. They're like, "Yeah, we wanna know how to make it better."

[00:31:19] **Aaron:** It seemed very authentic to me

[00:31:21] **Jethro:** Yeah. Well, the other thing about any AI company that exists right now, whether it's them developing the models or other AI companies that are creating tools for people to use, what they want is feedback. And they created this stuff in a certain way, and they use it for certain things, sure. But what they want is to see how other people actually use it and what their use cases are so they can, like, improve.

[00:31:51] **Jethro:** And, and one of the things, like plugins for example, is, like, really understanding what kind of plugins people want, and is it, is it better to do a plugin or to get an API key? Well, it's probably a lot easier and less technical to just do the plugin, and then you don't have to worry about a, a, an API key sitting out there somewhere that somebody else could get and accidentally put into code or whatever.

[00:32:18] **Jethro:** But, like, they want... They are so desperate for feedback that they will get on the phone with almost anybody and say, "What are you doing? How are you doing it? And let's figure out how to make it better." Because that's how people use this stuff. And so a, a question that I'm wrestling with now is, I don't know that Claude or, OpenAI's models are any better than each other's.

[00:32:45] **Jethro:** And I... In some situations, they seem like they are. In other situations, I, I just can't personally tell a difference at all. And most of the time, I can't really tell a difference. And so then I wonder, is this AI stuff, is it basically just going to become a commodity that these models are going to exist, and you just use whichever one you feel like in the moment?

[00:33:10] **Jethro:** And which eventually means you just use whichever one is the cheapest for you to use because they're all basically the same. And, did you get any sense of that from being there, that there was some, like, there is some secret sauce that makes OpenAI's better, or that they believe makes it better, or is it all just pretty much the same?

[00:33:33] **Jethro:** What, what's your current thought on that?

[00:33:36] **Aaron:** Yeah. I think something else a lot of attendees had in common, besides we're Codex power users, is we had been using Claude either primarily or a lot before getting into Codex. Almost everyone would compare things to Claude Code or Opus 4.7 to the new OpenAI model. I didn't get any sense from OpenAI people they are pushing for the narrative to say, but this is from attendees and myself personally Some common things that people pretty much agree on, design is not nearly as good as the best Claude models. So the, the top users would say things like, "Claude is really good at planning. Claude is really good at visual design," like typography and color scheme and layout of a website or an app. But I use it for those purposes, and I use Codex for these, and they're essentially working together. I think another common one that everyone agrees on, in my own personal experience, I've used Claude models a ton in the desktop app.

[00:34:47] **Aaron:** I've used Claude Code in the terminal for months. I'm on a Max plan. I don't use it as much anymore. Claude is not as persistent, and it's kinda like Claude is fun and nice and warm and happy and tells you what you wanna hear, but it's also lazy. And it'll do things like, "Oh yeah, I started that project.

[00:35:07] **Aaron:** We'll do the rest of it on Monday."

[00:35:09] **Jethro:** Yeah.

[00:35:10] **Aaron:** And this is, this is a quote from a developer. "I felt like I always had to babysit Claude, and it was down below me, and I had to like nudge it and get it to finish things. And that Codex with 5.5 is the first time I've had that can sit next to me more like a partner, and I can trust its work. follow through. persist on long-term tasks that run for hours." And it may not do the visual design as well, but you can have Claude give it the instructions. Or it may not understand the high-level planning as well, but have Claude write the plan and then have Codex execute it. Those were pretty, persistent and I've had the same. it's also unique 'cause Codex has a terminal, so on Mac, you just hit Command J, and you log into Claude Code, and then you say, "Yeah, use Claude for this." So I had CLI, Minimax, Claude Code all logged in, and then you tell Codex, "Besides sub-agents, you can use these other models for certain tasks."

[00:36:15] **Aaron:** Like if you need to generate music or video, use Minimax. When you're designing the visual layout of this website, use Claude Code Opus. And so these people are all way ahead of me, but they, they would have systems where they tie in other things. But Codex was essentially the, the, the main place they would orchestrate all the other models throughout.

[00:36:37] **Aaron:** One that no one can agree on is what about writing, like narrative, long form, turn my podcast

[00:36:47] **Jethro:** do you have-

[00:36:51] **Aaron:** I, I, I think it's just inherently harder to measure. my personal experience is that GPT 5.5 is less consistently good as Claude when it's good, it's spectacular, and when it's bad, it's terrible.

[00:37:05] **Aaron:** So it's kind of spotty and intermittent.

[00:37:08] **Jethro:** Yeah.

[00:37:08] **Aaron:** And Claude is, like, always solid and reliable and good at that type of writing. I have a few

[00:37:14] **Jethro:** Mm-hmm.

[00:37:14] **Aaron:** done hundreds of times with Claude and just dozens of times with GPT, but yeah. What about you? I mean, 'cause you've come from a Claude-centric world to figuring out ways to work in Codex and GPT.

[00:37:28] **Aaron:** What do you notice that it's better or worse at?

[00:37:31] **Jethro:** Yeah, so a couple things, that I have really noticed. What I love is that I can take a folder that is a GitHub repo, and I can just work in that folder, and it understands the folder and, and understands it well. I have been using OpenClaw and Opus... Or sorry, OpenClaw and Hermes for quite some time. And, and the closest I've gotten to that is with the little, Telegram groups with my OpenClaw that says, you know, "In this topic we're gonna talk about podcasting."

[00:38:13] **Jethro:** And I can not do anything in there, and then two weeks later say, "Use my Descript API to find this file," and then it does it, and it's awesome. And in fact, just a, a quick shout-out to Descript. Now that you can l- list the projects that you have, and I have thousands of projects, and w- before you could do that, it was basically useless.

[00:38:40] **Jethro:** Now that you can list them, it is absolutely amazing. And I did a recording, somebody asked for an old recording that never got published from October of last year, so well over six months ago. And I, I was like, "Oh my gosh, I'm gonna have to go find this. I do not wanna do that." And I just used the API to, to find it, and it brought it right up and oh my goodness, it was beautiful.

[00:39:07] **Jethro:** And then I said, "Publish it, and give me the URL here so that I can send it to this person who wants to review it." Oh, it was just beautiful. And, and I was using Codex for that, while we're on the topic here. But the thing is, and this is what makes me think that it's like, going to be a commodity very quickly, is that I, I used to use Claude for that, and then they cut off access to OpenClaw and said you have to use tokens.

[00:39:32] **Jethro:** And now I use, Codex for, For the, the access in OpenClaw and Hermes and, like, those kinds of things, what I don't want is have to go to five different tools to use something. I wanna be able to use pretty much everything in one place. I didn't know that I could orchestrate other models with Codex, so I'm gonna have to figure out how to do that.

[00:39:56] **Jethro:** I also haven't set up Telegram for, for Codex either, so I need to do that too. And that kind of stuff will make it so much better because I can't believe how much I can do through these agents like OpenClaw and Hermes that I can just say, "Hey, do this," and then it, it does it, and it has access to my computer with all the files there and takes care of things.

[00:40:22] **Jethro:** And it, it really is remarkable. And eventually we're gonna get to the point where we don't have to babysit those things as much, and they will just basically be personal assistants that are exceptionally good at what we need them to do, and, and that we're, we're quite close to that right now. And there's not a ton that I have to do to, like, keep things working.

[00:40:44] **Jethro:** So, you know, I, I will second your, your comment about Claude being better at design. Claude Design, that new thing is so good, and I have really enjoyed having that because I have always struggled with design myself. And in fact, my latest website, optimization doc is-- I designed all that with Claude Design, and then I told Codex to build that site based on what Claude Design gave it and, use the design files and all that kind of stuff.

[00:41:16] **Jethro:** And that, it, it looks exactly how I want it, and that's such an amazing thing, that I can have this vision in my mind, and then it comes to pass, when, when I ask it to do things, and that is, that is really remarkable to me. I definitely feel like we're living in the future.

[00:41:34] **Aaron:** And I'll say this about Claude Design, it's of all their products that people complain about rate limits, it seems to be the hardest to actually make it work for anything besides a one-off little project.

[00:41:45] **Jethro:** Uh-huh.

[00:41:46] **Aaron:** got this tip and used it for the first time days ago, but with GPT, if you don't have Claude plan the design, a sneaky way that you can make GPT way better at design, which I didn't believe, it sounded too simple. I had it use the built-in image generator, so GPT Image 2.0, which is pretty much the best model for anything right now. I said I was designing an interface, and I'm not a designer, but we all know, like, that's ugly or, you know, I know it when I see it. I had it generate an image of an interface for what I was building, and it looked amazing.

[00:42:24] **Aaron:** And then I said, let's build that," and it came out horrible.

[00:42:30] **Jethro:** Yeah.

[00:42:30] **Aaron:** tip that, this tip that, the tip that I got, which was... It sounds so simple, but they said, "No, no, no. Take the image and say-" /goal in Codex. It's like a Ralph loop. Make the actual interface functional, but look like this image you generated.

[00:42:46] **Aaron:** Don't stop until it works. And was like, "There's no way," because I already had it generate the image and then said, "Now make it." Just changing the, "Now make it look like that." And it took probably 35 minutes, which by the way, Claude models usually don't persist that long in my experience before

[00:43:03] **Jethro:** definitely not.

[00:43:04] **Aaron:** hit a tool call limit or have an error or something. It came back and the UI of this app, to the image

[00:43:12] **Jethro:** Hmm.

[00:43:12] **Aaron:** It worked. And I thought I could still use Claude, I still use it for some things, but that might be, since they have a built-in image generator in Codex, a sneaky way to have something else plan the visual layout and then say, "Now build that."

[00:43:30] **Jethro:** Yeah. Yeah, that's great. The other thing that, that is really terrible is if it doesn't get the image right on the first time, then it ain't getting it right. And I, like editing and improving, I have not gotten that to work at all when I've been using, any of the OpenAI tools. However, when I use edit in fal.ai, the ImageGen, pl- workspace, I have been able to get it to work there.

[00:44:01] **Jethro:** It can edit there, but it cannot edit for the li- to save its life in my experience, and I'm sure there's a prompting error that I'm making. But man, edits just do not work on images and... But if it does the one shot, then like you said, it is, it is an amazing image, and I've been very impressed with that for sure.

[00:44:22] **Aaron:** So for your use case, say it's a photograph of three people and your edit prompt is something like, "Remove this third person from the image," or,

[00:44:30] **Jethro:** Yeah.

[00:44:31] **Aaron:** to be this."

[00:44:33] **Jethro:** the lanyard says principal and it should say counselor, for example. Or, you know, it's, it's in a, it's outside of a school, but I want it to be in the school hallway, or any of those things. It just falls flat on its face, and it's better to just start fresh and in your original prompt change it.

[00:44:53] **Jethro:** That's been my experience at least, so if somebody has a different experience, I'd love to, to, for you to share your chat log and say, "This is how I got it to do it." Because that's the only way that I can, I can figure it out, is seeing what somebody else is doing.

[00:45:07] **Aaron:** Yeah. I haven't used GPT 2.0 for much editing of images, so now that you're saying that, I'm gonna have to try it

[00:45:14] **Jethro:** Yeah. And the good thing is, is that it is often good enough that you don't actually need to edit. Like, that's the other part that's amazing is I've had that happen numerous times where it is that good, and all of a sudden it's just exactly what I need, and that's awesome.

[00:45:29] **Aaron:** Yeah. Yeah.

[00:45:32] **Jethro:** All right. What else is on your mind? Anything else you wanna talk about from this experience?

[00:45:38] **Aaron:** No, I mean, I think, one of the overarching things is just a good reminder that when you put your opinions out there, whether it's your podcast, your videos, your tweets, your blog, your email, something, that good things can come from that. And, there are a

[00:45:56] **Jethro:** Yeah.

[00:46:01] **Aaron:** that don't get opportunities 'cause no one else knows what they're doing or they're not learning from them.

[00:46:06] **Aaron:** So it was a big reinforcing moment for that, which personally credit you for a lot of that mindset in myself as like, "Just keep tweeting," or, "Just keep posting," or, "Yeah, just make a podcast. It doesn't have to be a million download podcast to be worth doing." That's been, that's been very true. And I think it's also reinforced, one of my biggest takeaways is such a gap between the bleeding edge, "I know how to use all this stuff and I'm building crazy things." For example, there's a guy that owns a construction company in Arkansas, and has no project managers. It's all AI agents with OpenAI stuff. And I didn't

[00:46:50] **Jethro:** Yeah.

[00:46:54] **Aaron:** business and one of the main manager level positions doesn't exist in his company, and he's used AI to do it. And I go, the gap from him to the average construction person, which is like, "I asked ChatGPT if the grocery store was still open," it's like a Google search replacement, is getting wider every day. And what I've found to be true for myself is I don't need to be the bleeding edge technical, get a machine learning doctorate. I can just be somewhere in the middle and understand a little bit of the bleeding edge side of the spectrum, but actually communicate back and bring along people who cannot understand the bleeding edge. They just need a simple practical use case. I tell people this all the time. I think that's actually the space you wanna be in today is somewhere in the middle.

[00:47:45] **Aaron:** Like m- when I characterize what I do, I tell people I'm the conduit between the average person and the really techy person. I'm neither of those, but I can talk to both. As the gap gets wider, there's more room and more need for middle people like me I think you would say the same. You're not a computer scientist, but you understand these things well enough to explain it to the average business owner or the average principal or the average podcaster.

[00:48:13] **Jethro:** Well, and, and what, where I'm at right now is I understand the tool well enough to get the tool to do something for me, but I don't understand it well enough to make it itself, right? So I don't know how to create a new model that's going to do very specific things, but I do know how to get the current models to do things that are instantly applicable and meaningful to normal everyday folks.

[00:48:42] **Jethro:** And, and as I have-- Over the last week, I've kind of handheld a couple people to, to do something that they haven't done before. Like on a Zoom call, they're in Claude Code for the first time, and I'm saying, "Do this, do that. Okay, just try it and see what happens." And both of those people said, "After meeting with you, I went and just did it on my own and just like went to town and felt totally confident and comfortable that I could make it happen."

[00:49:14] **Jethro:** And I do think there's real value in that middleman like, "Let me hold your hand. Let me teach you. Let me like approach it as a teacher, not a, not a salesperson, not a course creator or something like that, but let me approach it as, as a teacher and help you feel confident in doing it." And teacher is the wrong word here because teacher typically means somebody standing at the front of the classroom explaining everything.

[00:49:42] **Jethro:** I mean like the actual teachers that are really amazing, which is, "Let me help you actually learn and apply something, not just put information into your brain." Because that, that kind of a teacher is the one that we need, is, is somebody who will help you get confidence, feel good about it, and move forward.

[00:50:01] **Jethro:** And that's incredibly powerful and empowering for people so that once you do that, they're like, "Okay, I'm off to the races and I'm gonna do something else with this." And, and that's where that middle position that you're talking about becomes really valuable for the people who are receiving that because then they can, they can move towards that high end, as far as they're comfortable and, and that's a really valuable thing.

[00:50:30] **Aaron:** And what I tell people is you wanna have some group or person ahead and behind where you are with these things. Where you

[00:50:37] **Jethro:** Mm-hmm.

[00:50:38] **Aaron:** Maybe Jethro's ahead of me. We talk on Zoom once a week and share tips, and I learn from him. But I'm also teaching somebody, even informally, like my spouse, my kids, my coworker who's a few steps behind, I might think I understand it, I did it for myself, but could I explain it to someone who has no concept of what that is? And increasingly, I find value, and I know you, you have these places too, in these group chats, these Discord servers, these regular meetings you have with people where it's, it's not pay to play, it's just nerds sharing and asking questions. And then you're like, "Whoa, this is beyond anything you could get on YouTube or LinkedIn.

[00:51:20] **Aaron:** No one's talking about this." And those places exist, and that-- I actually get more value from those types of sources than I do anything that's formally like a course online or a YouTube video that somebody puts out on these things.

[00:51:36] **Jethro:** Yeah, because we're still humans, and we still go back to who we are as human beings, which are people that s- that thrive in communities and thrive when there's connection. And so when that's the case, and those, those places that you're talking about, those do exist and, and there's real value there. And I also go find pay-to-play pa- places so that I can be in the same room as other people who are, you know, even if they, they rarely say anything, I at least have a connection with them and can...

[00:52:14] **Jethro:** You know, somebody will, will share one thing and I'll be like, "Oh my goodness, my brain is exploding because of that thing that they just, like, threw out there as, like, a nothing comment." And I'm like, "What? That is awesome. That's exactly what I've been looking for." And that has happened to me so many times in my life that it is worth it to be in the same room as high-quality people, and, worth it to be just in their presence, and to be that person for other people as well.

[00:52:44] **Jethro:** And, you know, going back to that same thing that we've been talking about, that's why it's important to share, to talk about what you're doing, to not keep it in the dark. And there's, there's value for you in sharing it and in, in, in hearing what other people share as well.

[00:53:01] **Aaron:** Yeah, absolutely. Well, I appreciate you letting me be a part of the sharing today, Jethro.

[00:53:07] **Jethro:** Yeah, this was good. how do you want people to reach out to you, get ahold of you, see what you're working on?

[00:53:14] **Aaron:** Yeah, I mean, the website's aaronmcelky.com, but you can follow on your drug of choice when it comes to social media platforms. I try to, I try to post stuff everywhere. But yeah, website has my email and, LinkedIn and Twitter are probably the two main ones these days

[00:53:33] **Jethro:** Yeah. Very good. Excellent. Thanks so much, Aaron. Always a pleasure. to be honest, I don't even know what podcast I'm gonna put this on. I might put it on a whole bunch.

[00:53:43] **Jethro:** All right. Welcome to whatever podcast you're listening to. I'm Jethro Jones, and today we have Aaron McKelky, who is,