Square note that in hospitality, efficiency is everything. That's why the system lets you take payments, track sales, handle inventory, manage staff, send invoices, and keep up with finances all in one place, fly through orders with zero mistakes. Get the data you need and keep everything working together. So you're ready for whatever's next.
Learn more about their customizable plans at squareup.com. Hi everyone, this is Pivot from New York Magazine in the Box Media Podcast Network. I'm Tara Swisher. And I'm Scott Jallaway.
And you're listening to our special series on AI. We talk a lot about the business of AI, but today we want to focus on the ways we can actually be using it in our day-to-day lives. But here to chat with us about some of the AI basics is Kylie Robinson. Kylie is a senior AI reporter for The Verge.
Welcome, Kylie. It's good to talk to you. Yeah, thank you for having me. So first of all, talk a little bit about how you personally use AI in your day-to-day life.
And why are you covering it? Yeah. Before this, I was covering Twitter at Fortune Magazine, which you were run by Cara Swier person. And that wasn't easy to cover.
And I thought AI showed a lot of promise, and it was just such an interesting area to tackle for a young reporter. I mean, who doesn't want to be covering the biggest technology to hit the scene. And I live in San Francisco, so it just seemed perfect. But it is perhaps the most stressful beat I've ever had, because it's so large and so nuanced.
And people argue about it all day. About using AI in my day-to-day life, I would say it's not something I use heavily. I do use it for... I upload a document, for instance, OpenAI will release these safety cards for their models that show, like, this is how safe it is.
So I can upload that PDF to GPT-40 and say, okay, and ask questions based off that PDF. Do they mention this? Can you expand on what this means? So like, really, heavily technical documents are white papers.
I can ask questions and simplify it in a way that's more helpful. It's quicker for me to understand and then going to a bunch of researchers, making a bunch of calls to get them to explain it. I think that's been really helpful for me. I know Scott uses it for writing.
I think... I've noticed Claude can be kind of a helpful writing assistant, but in terms of actually using it to write, I don't use that because I don't think it's helpful yet. But just as a partner, as Scott has mentioned on the podcast, as a partner, to be like, okay, here are some of my rambling thoughts. Can you streamline what I'm trying to say and edit it for me?
Right, absolutely. So give us a picture of someone who's is looking to implement AI in their lives, like daily tasks that can be made easier when people ask you this question. Obviously, Google is integrated into writing emails, for example, which I don't find useful, but dills, resumes, and much of what the average person knows of AI is chat, GBT, and related tools. Try to expand on that.
What other tools are useful for them? Yeah, there's so many different AI tools now. I think a lot of people use grammar leads so that can check your grammar in your browser, which is really helpful. I think when you want to use AI, I think you should consider it for low stakes tasks.
You have to imagine your data privacy because often these models will use what you input to train the model. So you don't want it someday spinning back your bank account information, which is really funny because I'm a big listener pivot and hard fork, which Kevin, one of the hosts there, uploaded his bank statements to Notebook LM, so they could create a podcast to help him with his financial information, which is a really interesting thing that I think people want AI to be capable of right now is that can you help me budget, which I think, again, low stakes tasks, thinking of the data you upload, trying to upload sensitive information, I think just like I said with the PDF, that was really helpful. I just turned 26, so I used it to compare health insurance. These are my needs.
These are the options I have. Here's a PDF of what they offer, which one should I choose was really helpful, stuff like that. Nice to meet you, Kylie. By the way, I use it this morning.
I asked AI, why am I so broke? And it immediately sent me a copy of an email confirmation from Amazon that my Mexican cat costume will be delivered tomorrow. It's a little AI humor, Kylie. I'm so sorry, Kylie.
It's a young woman. You need to endure this, but here you are. All right. I'll say, how can I feel better about myself and it'll just come back with good morning, you fucking stack us on shine.
So anyway, a lot of people use AI, a lot of people having been started. What would you suggest someone does to get started? Which one or two LLMs would you suggest they download, the free version even? And how did they start to try and unlock the potential and just understand it more?
And of course, I'll start with an answer. The first time I really interfaced with AI was trying to figure out fun stuff for me and my fortunate son to do in London. And it kind of went from there. What two or three things would you suggest to help people get started?
Yeah, I think that's a really helpful example and something I think the makers of these models want people to be using it for rather than highlighting any nefarious ways to use AI. They're helping people use it to write a story for their young child, set with pictures. You can use ChatchPT for that or Grock. And you can use it to book track or like to plan your travel, which is really cool.
And I used it just when I went to Spain, I was like, what should I go see? So I think those are really helpful examples. Again, these are all low sticks tasks that you can use any really any chatbot on the market that's capable enough to use creatively. I think you have to check to see if these places are open and exist because it can hallucinate very confidently.
But yeah, any model right now, they're all kind of on par with each other. A lot of them are because they're all working towards the same thing. So, you know, you can use it to decide what hair color you want to do next. Anything that's just fun and low sticks, I think is easy.
Low sticks. So you did mention privacy. Now we do put a lot of privacy online that my bank things are online, everything else. But I really limit what I use here because of that because I'm like, we're very worried.
And I'm somebody who's very aware of privacy. I'm like, Scott, when you said you put your record, your medical records, I'm like, I'm not putting my route into the open AI in a way. They already know Cara, your privacy is gone. Yeah, I guess, but I just don't want to help them along to like put all the things together.
And so I just they definitely don't have my heart surgery stuff. They don't. They don't. And so you're worried about the CCP, like scaring you to give your heart attack?
I don't know. I just I'm just telling you the feeling I had. Kylie, this is what I deal with. This is what I do with it.
This is my feeling like I don't want to give him too much personal information. I don't mind it on writing things and like there are low stakes. But how do you but then now we put lots of stuff on the regular night. When do you imagine that crossover for the how should the average person feel about that?
Because we don't know where what this stuff is being used for, correct? Like there's not as much transparency as there needs to be. No, there's not as much transparency. And they claim that it's because of like Scott said, the CCP or anti-competitive reasons.
I think they've already hoovered up the entire internet. There's no going back from there. All these models have hoovered up the entire internet. It's something I've kind of reckoned with when they said, you know, photo bucket signed over all of its data to train large language models.
I was like, dang, like whatever I put on photo bucket, that was stupid when I was 11 is now going to be used to train large language models. And 11 year old me wouldn't have known that. So I think it's kind of, it's a really tough position. People on Instagram, celebrities were like, you know, meta does not have the right to use my photos if I post the story.
I think people are really protective over, you know, the lives that they have shared with the internet that they have been encouraged by these large companies to share with the internet. And it's just, I think they feel like it's being taken away from them and, you know, used to train these black box models. I think people have different opinions on it. I personally feel I have the eGBs about it because I have, you know, I grew up with the internet, with Facebook launching when I was a young team.
So I think it's a very tough position. I think some people are like, I don't care. So do you think that people are more worried because a recent study found that one in nine Americans use AI every day in the world? That's a very small number.
Like at this point, right? Where are we in that? Do you think everyone's just going to do it? Like, I'll not do it.
It's going to be done to them. And that'll just be foisted upon the bi-apple intelligence or whatever. You know, do you want to get a new but you have your airport? That seems like a good thing, for example.
Totally. I think automating, you know, wrote tasks is not a bad thing. When it comes to the workplace, I think a lot of workplaces are well aware of the data privacy issues. And they're like, please don't upload our internal documents to open AI.
That's been a problem. It's trained in a lot of workplaces. So one in nine doesn't surprise me. I think it's going to continue to grow.
I just published a story today about AI agents, which is just like the new next thing. So sort of an AI assistance. And where I'm seeing this a lot is in SAS products, like Salesforce released a CRM agents, Microsoft has co-pilots, stuff that they believe will increase efficiency amongst their staff. But I think it's going to be hard for that number to grow so long as there's transparency issues and that trust has to grow.
Okay, let's take a quick break. We come back. We'll talk about where we're already using AI without realizing it and what we should not be using it for. Some say the bubbles in an air travel piece can take 34 seconds to melt in your mouth.
Sometimes the very amount you're stuck at the same red light. Rich, creamy, chocolatey air travel. Feel the arrow bubbles melt. It's mind bubbling.
Amazon presents Jeff versus taco truck salsa, whether it's Verde, Roja, or the orange one. For Jeff, trying any salsa is like playing Russian roulette with a flamethrower. Luckily, Jeff saved with Amazon and stocked up on antacids, ginger tea and milk. Hebiniero, more like habinier, yes.
Save the everyday with Amazon. Scott, we're back with our special series on AI. We're talking to a senior AI reporter for the Verge, Kylie Robinson. Huge leaps have been made in AI over the last couple of years.
Talk about how we're using it without realizing it. It's also been around for a while. We're people not realizing they're using it today. Totally.
I think in your career care, you've probably covered AI. It's been around forever. I think your Netflix algorithm that's AI, automated self-driving cars, Weimos, I live in San Francisco. Weimos are everywhere.
That's AI. It is used in TikTok algorithms. That's AI. It's everywhere.
It has been working in the background quite a bit. I think you'll hear companies, especially as a reporter, they're like, we've been in the AI business for two decades, which is not facetious, but it is different than the frontier models we're seeing. Open AI and anthropic release, but it is those algorithms you're used to using. So break down or do a kind of JD powers review of the biggest LLMs from your favorites, your least favorite?
Favorite to least favorites. Scott likes Claude. I do like Claude. Claude is really good.
It's surprisingly good. I just started using it recently and I messaged a co-worker. I was like, I'm a bad AI reporter because this is way better than I anticipated. It also has, I don't know if you've noticed, it's got like kind of intense guardrails.
I asked some questions about AI. It's like, well, as an AI, I can't exactly answer these questions, whereas chat to PT would have just spit it out. So people know it's why it's why anthropic, which was a group of people who thought Open AI was not safe enough and started anthropic. Exactly.
And it's backed by Amazon. Exactly. Yes. And Google has a smaller cloud share for anthropic, but yes, so anthropic is a competitor to Open AI.
Open AI has GBT4O as their latest frontier model. They've also released a reasoning model called 01, but they consider that for lack of a better word, kind of dumber than their frontier model. And frontier models are basically the biggest, the best, you know, models that are out there. So frontier models are like, the next one is like the next iPhone, basically.
So I would say Claude is amazing. Claude Opus is amazing. I think, I think the thing is they're all building the same thing with the same training data, which is the entire internet. So they're going to continue to leapfrogging over each other.
So it's hard to compare it because it's, you know, five major companies with some of the best researchers in the world with all of the same training data, all building the same thing. Do you use Grok? Do you use Grok? I'm not getting in any of the new, whatever cyber attacks again, I'm not getting in and I'm not putting any information into any of his properties.
I don't trust him personally. I use Grok when it first came out to, I think, I made, you know, Kamala with a gun is, you know, the bridge put out a story because they have Grok for the for the listener is available on X, formerly Twitter, which is on by Elon Musk, and he owns XAI, which created this chat bot, and it has what it feels like, no guard rail. So you can make a lot of photos that, you know, break all sorts of copyright laws. No, I don't use Grok and I don't necessarily find it to be top of the line.
If I were to rank models, they're not, you know, at the top. So I'll go back to my question. What's your favorite LLM? I would say Opus.
Claude, it's really intelligent and incredible. I think, you know, it's enviable from other labs what they built. And then what, can you name some long tail LLM or AI apps that are sort of fun that people that maybe haven't gotten very much attention that are kind of fun. Any sort of undiscovered gems out there?
Undiscovered gems. I mean, if you go to Huggingface, if you're into open source, there are hundreds of thousands of open source LLMs that people can mess with. I mean, that's the cool part about open source LLMs, which is a very hot debate, but, you know, developers are creating all sorts of cool shit with the open source LLMs available on Huggingface. So there's almost too many to choose from, but none of them are mainstream in the way because it costs so much money.
Hence why opening I just raised the most money that anyone's ever raised ever. It's a lot of money. It's just for people in Huggingface, it's an AI community where it's a platform where they collaborate and they do different things. And, you know, this will be very much like the early app days or the early internet days where there was suddenly websites and things.
And then there was Yahoo that compiled them into yet another hierarchical, vicious oratory. I have a follow-up. I find that AI is very politically correct that it will say, to answer this question, you know, make sure that you check with law enforcement or- That's not politically correct. Oh, I find it very politically correct.
Please don't steal from the jewelry store. All right. No, but it'll come back and say, well, this might reflect bias or you should. I just find it's very- I'm looking for an AI that will say, that's a stupid fucking question.
Or your question, I want it as a friend to say- He wants to shame AI. Yeah, thank God. Hit me harder. Call me daddy, you bitch.
Oh my God. No, but I do find it's very- They're so worried about it going weird places that it's constantly preconditioning and qualifying all its answers and being very gentle. I find it's very overly sensitive and quite frankly politically correct. So I'll start with Kylie.
Do you find that to be the case or do you think that's just putting inappropriate guardrails? I think they're putting inappropriate guardrails because it's so nascent. I mean, why start off crazy? I feel like we can work our way up to getting you a sadistic chatbot.
But for now, it's so- Go on. It's just such a nascent technology. So I think being overly safe and correct and nervous about what it's going to output to millions of people, I think that's a good move. Yeah, Scott, come on.
I'm going to answer something. I know you want to please, bitch, AI. But one of the things that's really important is it doesn't sexually harass people. It doesn't- Okay, you're taking this to an even darker place than I would go.
But I'm saying it has. It's been the original ones were racist. If you ask it a simple question, it'll start conditioning everything and telling you to check with us and make sure that you talk. And it's just sort of just give me the goddamn answer.
I get that. But they're never going to do that because they literally, the first time they put out some Microsoft stuff, it was racist, right? It started to say racist things. So they really can't have one of the things that I think I tell a lot of people is I met these two guys on the street yesterday and they were creating an AI.
They just ran up to me. They went pivot. They're creating an AI that goes on speaking of unusual things that goes on top of 911 calls that they'll be selling into cities. And it'll translate say Spanish immediately because not every person, there's a delay there because the person who's taking calls not Spanish and they have to go get a Spanish-speaking dispatcher.
And so it's doing all kinds of things that Groxon sends things out really quick. I thought it was a great idea. That was a really interesting way to use an AI. And I said, but you know what?
Something is you can't make a mistake even if human-ditch patches do. So I think they have to be unusually careful with all these things as this AI is shoving us around the planet. I don't know. I just feel like that's okay.
You can take it, Scott. But I'm going to make, I'm going to make you a mean AI overlord or please bitch or something like that. I learned this morning from AI that a group of flamingos is called a flamboyance. That's true.
And I love that. I love that. That's awesome. In the dictionary.
Okay. Here are flamboyance. Anyway, last question. We just sort of covered the idea of what things we should not use it for.
It's going to use for everything, just FYI. But any predictions, last question on things AI can't do yet, but we'll be able to do for us in the next three years. Use your imagination hat here or things you're seeing or hearing. I think in the next three years, again, these are so hard to tell because you need so much money and so much compute.
So if we could just continue on that exponential curve that these companies are hoping for, I see probably more accurate and natural voice interactions. That's something that they're building, that they really want the her movie-style reality. I do think that will get better, especially as they release all of this to the public and people test it and they train on people using it. I think those will naturally get better.
Advanced code generation and debugging, that's something they're already really good at. And if these reasoning models from open AI and others continue to get better, it's going to be better at coding and debugging, which will be really cool. And they're all building agents, which again are like little AI assistants. That's sort of the high stakes tasks that they want to access.
Hence why all these guardrails are so tough because they want these high stakes tasks like running your life and booking you flights and having access to all of this. So I do see them building out agents, but it would require so much compute. And so much money to get there. So I'd be curious what you guys think because I get asked all the time, like, is the bubble going to pop?
Is the bubble going to pop? Is opening AI just going to crash? Which I think it's so hard for me to tell you guys have been doing this for so long. It will, but no, no, no.
It's like when the internet crashed, this is a big deal. This is a change in computing. It's yet another great change in computing. This is not crypto.
This is not some of the little bubbles, a bubble, I guess, but it's directionally correct. It's directionally, and it's going to be huge and encompass everything. Scott? Well, there's two things.
There's evaluation of these companies, and then there's the real impact they have in the economy. I think the latter is just getting started. There's going to be a lot, it's going to be what I would say in terms of evaluations. There's going to be a lot of volatility.
We've talked about this. We think that relative to its size and leadership position, open AI, in my view, is actually at 12 times revenues, is actually probably the best value because some of the long-tail ones you talked about who have almost no revenues and no real visible business model yet still get to $10, $20, $50 billion valuation. So it's going to be a wild ride. I would describe it as like late 90s internet.
We don't know if it's $97 or $99, but we know that by 2005, it's going to be much bigger than it is now. That's a long way to wait. Kiley is saying, I have no idea. No, he does.
It's up and to the right eventually. Anyway, thank you, Kiley. We really appreciate it. You can read Kiley on the verge.
It does amazing work on this topic and breaks a lot of stories. A colleague. A colleague. She's a Scoopster.
She's a Scoopster and she's a great one at it. Anyway, okay Scott, that's it for our AI Basics episode. Please read us out. Today's show is produced by Larry Name and Join Marcus and Taylor Griffin, earning your Todd Engineer this episode.
Thanks also to Drew Burroughs and Ms. Lderio. Ms. Chakra is Vox Media's executive producer of audio.
Make sure you subscribe to the show where every listen to podcasts. Thanks for listening to Pivot from New York Magazine and Vox Media. You can subscribe to the magazine at mimag.com slash pod. We'll be back next week for another breakdown of all things tech and business.