--:-- Sean Rameswaram: Welcome to Today Explained. In this episode, we'll be discussing the topic of AI generation. This refers to the use of artificial intelligence to generate various forms of content such as text, images, and audio. AI generation has the potential to revolutionize many industries and has already begun to change the way we create and consume content. We'll explore the implications of this technology and discuss some of the challenges it presents.
--:-- [Sound effects: Upbeat music starts playing]
--:-- Sean Rameswaram: Support for this show comes from ServiceNow. AI is only as powerful as the platform it's built into. That's why it's no surprise that more than 85% of the Fortune 500 use the ServiceNow AI platform. While other platforms duct tape tools together, ServiceNow seamlessly unifies people, data, workflows, and AI, connecting every corner of your business.
--:-- James Vincent: [Voice of a guest] Hi, I'm James Vincent, senior reporter at The Verge. I'm here to talk about the latest wave of AI-powered creations.
--:-- Sean Rameswaram: The internet's been full of screenshots of people's DMs lately, but they're not DMing with other people. They're messaging with artificial intelligence. The internet's going crazy over new artificial intelligence called ChatGPT.
--:-- James Vincent: Yeah, it's an AI chatbot that's made by a company named OpenAI, who have been doing a lot of research into this and other AI areas. And it's sort of the most, one of the most advanced versions or one of the most advanced examples of its type.
--:-- Sean Rameswaram: But the big thing is, is that it's sort of publicly accessible in a way that these level of bots, these high-quality bots haven't been before.
--:-- James Vincent: So, you know, you can just jump into a conversation with it straight from your browser, open up a window, you know, like you were talking to a family member, and you can ask it questions, you can ask it anything you like.
--:-- Sean Rameswaram: And it's, it's pretty clever. How does it work?
--:-- James Vincent: ChatGPT, like other sort of AI chatbots, is what we call a large language model. And they are trained by looking at huge stores of text, most of which have been scraped from the internet.
--:-- Sean Rameswaram: Can we try it out right now? Can we play with it?
--:-- James Vincent: Absolutely. Yeah, we can load it up, have a little talk.
--:-- Sean Rameswaram: Where do I go? GPT.org, chat.openai.com forward slash chat.
--:-- James Vincent: Okay, that was my second guess, chat.openai.com.
--:-- Sean Rameswaram: Oh, I need to log in. Do I have a login?
--:-- James Vincent: Uh, you may need to sign up.
--:-- Sean Rameswaram: I've got a login with me now. Would you like me to... I'm all ready to go if you'd like me to ask it some questions.
--:-- James Vincent: Okay, why don't you just do it and I'll just... I'll just be entertained.
--:-- Sean Rameswaram: I want to ask it something more nuanced. Like, why do people love Beyonce?
--:-- James Vincent: Okay, let's see what it says.
--:-- Sean Rameswaram: There are many reasons why people love Beyonce. She is an incredibly talented singer, dancer, and performer, and has been entertaining audiences for many years.
--:-- James Vincent: That's not wrong as far as answers go.
--:-- Sean Rameswaram: I could see if this were like a fourth-grade, you know, class and we were doing an essay, write a paragraph about why people love Beyonce. That would be a pretty good answer from an eight-year-old.
--:-- James Vincent: But let's try and push it a little bit further then.
--:-- Sean Rameswaram: Are you a Beyonce fan?
--:-- James Vincent: Sure, sure.
--:-- Sean Rameswaram: Okay, so what's something a little bit more granular? What's your favorite song by Beyonce?
--:-- James Vincent: Ooh, that's tough.
--:-- Sean Rameswaram: Wow, I didn't know that you were going to be putting me on the spot.
--:-- James Vincent: I mean, sorry.
--:-- Sean Rameswaram: No, it's okay. I like it. I appreciate it. Plastic Off the Sofa is a recent favorite.
--:-- James Vincent: Okay, so why don't we ask the bot, Tell me why plastic off the sofa is such an amazing song?
--:-- Sean Rameswaram: So there may be a problem with this because the chatbot is only trained on data that goes right into 2021.
--:-- James Vincent: So you've picked a recent one and it said, I'm sorry, I don't have any information about a song called Plastic Off The Sofa.
--:-- Sean Rameswaram: So you've stumped it already. We found its blind spot.
--:-- James Vincent: But let's ask it to be creative. What if we said, write me a song in the style of Beyonce?
--:-- Sean Rameswaram: Could it do that?
--:-- James Vincent: Yeah.
--:-- Sean Rameswaram: What should the song be about?
--:-- James Vincent: Love.
--:-- Sean Rameswaram: Come on, that's too easy. Too easy.
--:-- James Vincent: Let's have something that Beyonce wouldn't write a song about.
--:-- Sean Rameswaram: Because one thing these bots are very good at, they're very good at combining styles.
--:-- James Vincent: Okay, great, great, great. Write me a Beyonce song about cheeseburgers.
--:-- Sean Rameswaram: All right. Okay, verse one.
--:-- James Vincent: I mean, I'm reading this out loud, so this is going to be terrible.
--:-- Sean Rameswaram: We'll get you a beat. We'll drop the beat for you.
--:-- James Vincent: I'm in the mood for a cheeseburger. Something savory, something sweet. Juicy patty, melted cheese. Topped with ketchup, lettuce, please.
--:-- Sean Rameswaram: Cheeseburger, cheeseburger. It's my favorite food in the whole wide world.
--:-- James Vincent: Cheeseburger, cheeseburger. I would eat one every day, girl.
--:-- Sean Rameswaram: It's not terrible.
--:-- James Vincent: That's even worse in my accent.
--:-- Sean Rameswaram: I really apologize for that.
--:-- James Vincent: This is the most un-Beyonce thing anyway.
--:-- Sean Rameswaram: Oh man, that's great.
--:-- James Vincent: Okay, so it's creative. The AI can do creativity.
--:-- Sean Rameswaram: Absolutely. And it can do, you know, all sorts of styles of writing.
--:-- James Vincent: So, for example, if you wanted to, say, write a college essay or you wanted to write a legal document or you wanted to write a news article, you could ask it to say, write me a news article about landing on Mars.
--:-- Sean Rameswaram: And we can put that in there and then it'll come out in a few seconds.
--:-- James Vincent: Seconds? A few.
--:-- Sean Rameswaram: Yeah. I mean, it's not too slow.
--:-- James Vincent: Oh, sorry.
--:-- Sean Rameswaram: It's written me another song about cheeseburgers.
--:-- James Vincent: Sorry about that.
--:-- Sean Rameswaram: Well, who else would you like to land on Mars, apart from Beyonce? Who else could be landing on Mars in the future?
--:-- James Vincent: Let's say, write us an essay about Matt Damon landing on Mars.
--:-- Sean Rameswaram: They say that once you grow crops somewhere, you've officially colonized it. So, technically, I colonized Mars. Matt Damon.
--:-- James Vincent: Okay. Landing on Mars. It's thinking.
--:-- Sean Rameswaram: Actor Matt Damon has made history by becoming the first person to land on Mars.
--:-- James Vincent: The Academy Award-winning actor was part of a NASA mission to explore the red planet and successfully touched down on the surface earlier today.
--:-- Sean Rameswaram: Damon, who has long been fascinated by space exploration, was selected by NASA to be the first person to land on Mars due to his experience as an astronaut in the 2013 film The Martian.
--:-- James Vincent: It made the connection! It made the connection.
--:-- Sean Rameswaram: See, it is the little things like that that are very impressive about this.
--:-- James Vincent: Because you're like, well, I never made that explicit. That was implicit within your statement. And it found it.
--:-- Sean Rameswaram: In your face, Neil Armstrong.
--:-- James Vincent: So the AI can give me a song about cheeseburgers in the style of Beyonce. The AI can write me an essay for a future news event and connect some cultural references that I didn't even make explicit.
--:-- Sean Rameswaram: What are the limitations here? Where does this really show its gears?
--:-- James Vincent: So there's two big limitations, one of which is sort of hard to see on demand. And it's the fact that these systems will make up information if they don't have a good answer for you.
--:-- Sean Rameswaram: So they really have no way of telling fact from fiction.
--:-- James Vincent: It's not like an encyclopedia where it has a list of known things that happened. It's a probabilistic machine. It's only predicting the probability of the next word being correct.
--:-- Sean Rameswaram: Sometimes you might ask it for a biography of someone. And if it doesn't know what they were doing between the years 73 and 79, it'll just make it up.
--:-- James Vincent: And the probability stuff means that it's also bad at some things like maths.
--:-- Sean Rameswaram: So if you ask it to be like, oh, what's three times three, that is something it will have seen in its training data.
--:-- James Vincent: So it'll know what that is, or it'll know enough sort of small sums to know where that's going.
--:-- Sean Rameswaram: If you ask it to multiply two very large numbers, the chance of it having seen those numbers anytime before is next to nil.
--:-- James Vincent: So it won't know how to do that.
--:-- Sean Rameswaram: Wait, it can't do large multiplication? That feels like so basic, like my phone could do that.
--:-- James Vincent: But this is like the promise of this technology, right? Is that it has these huge strengths.
--:-- Sean Rameswaram: And it's because we are approaching how to make computers in a completely different way to normal.
--:-- James Vincent: It is built on these principles of machine learning, which is all about learning from past examples.
--:-- Sean Rameswaram: So it's created this sort of statistical map of how language works and how text works.
--:-- James Vincent: And it's done that by looking at lots of texts great from the web.
--:-- Sean Rameswaram: But that means that its knowledge is probabilistic rather than deterministic.
--:-- James Vincent: So it only knows the probability that the next word will follow will be the one you want.
--:-- Sean Rameswaram: And it doesn't have a sort of a system of hard-coded facts like an encyclopedia might.
--:-- James Vincent: Let me hear you say hey, Miss Carter.
--:-- Sean Rameswaram: The ethical issues because obviously there are ethical issues in a minute on today explain.
--:-- James Vincent: I'm in the mood for a cheeseburger something savory something sweet juicy patty melted cheese top with ketchup lettuce please cheeseburger cheeseburger it's my favorite food in the Whole Wide World. Cheeseburger, cheeseburger. I would eat one every day, girl.
--:-- Sean Rameswaram: Support for Today Explained comes from Chime. Nobody likes fees and complicated banking, says Chime.
--:-- James Vincent: And Chime gets that, says Chime.
--:-- Sean Rameswaram: Chime is a financial technology company that says they understand that every dollar counts.
--:-- James Vincent: So when you set up direct deposit through Chime, they say you get access to fee-free features such as free overdraft coverage, getting paid up to two days early with direct deposit, so much more.
--:-- Sean Rameswaram: Chime says with qualifying direct deposits, you can be eligible for free overdraft, up to $200 on debit card purchases and cash withdrawals.
--:-- James Vincent: Plus, they say you don't need to worry about monthly fees or maintenance fees.
--:-- Sean Rameswaram: You can work on your financial goals through Chime today.
--:-- James Vincent: You can open an account in two minutes at Chime.com slash explained.
--:-- Sean Rameswaram: That's Chime.com slash explained.
--:-- James Vincent: Chime feels like progress.
--:-- Sean Rameswaram: Botany eligibility requirements and overdraft limits apply.
--:-- James Vincent: Timing depends on submission of payment file.
--:-- Sean Rameswaram: Fees apply at out-of-network ATMs.
--:-- James Vincent: Bank ranking and number of ATMs according to U.S. News and World Report 2023.
--:-- Sean Rameswaram: Chime checking account required.
--:-- James Vincent: We all have moments where we could have done better.
--:-- Sean Rameswaram: Like cutting your own hair.
--:-- James Vincent: Oof, yikes.
--:-- Sean Rameswaram: Or forgetting sunscreen so now you look like a tomato.
--:-- James Vincent: Ouch.
--:-- Sean Rameswaram: Could have done better.
--:-- James Vincent: Same goes for where you invest.
--:-- Sean Rameswaram: Level up and invest smarter with Schwab.
--:-- James Vincent: Get market insights, education, and human help when you need it.
--:-- Sean Rameswaram: Learn more at schwab.com.
--:-- James Vincent: All the single patties.
--:-- Sean Rameswaram: All the single patties.
--:-- James Vincent: All the single patties.
--:-- Sean Rameswaram: All the single patties.
--:-- James Vincent: All the single patties.
--:-- Sean Rameswaram: All the single patties.
--:-- James Vincent: Get your cheese out.
--:-- Sean Rameswaram: Today Explained, we're back with James Vincent, senior reporter at The Verge.
--:-- James Vincent: We talked about the AI-powered chatbot that's been passed around online, but it's not just text, it's images too.
--:-- Sean Rameswaram: Yeah, they're related in that they're both part of the same sort of genre of AI called generative AI, which is all about sort of replicating human creativity, basically, whether that's writing things or drawing things.
--:-- James Vincent: And they also use some of the same sort of underlying technical methods.
--:-- Sean Rameswaram: You know, they both involve looking at large amounts of data and then looking for patterns in that.
--:-- James Vincent: And we've, you know, we've come up with some very good ways of doing that in the last couple of years, which is why these technologies are sort of bubbling up at the same time.
--:-- Sean Rameswaram: Is this the same chat bot, a different bot?
--:-- James Vincent: These are different bots, but some of which are made by the same company.
--:-- Sean Rameswaram: So there's been a lot of image generation bots that have become really popular in the last year.
--:-- James Vincent: One of which is called Dolly, which is also made by OpenAI, like ChatGPT.
--:-- Sean Rameswaram: But there are also sort of very open source ones called Stable Diffusion, Mid Journey, and they're free for anyone to use.
--:-- James Vincent: And in the past, these systems were sort of quite inaccessible. They were mainly used by researchers, but now they're available on the open web.
--:-- Sean Rameswaram: And so people are asking them to do all sorts of weird things.
--:-- James Vincent: Yeah, I mean, I saw this one guy who I follow, Hank Green, who's a friend of the show, say, I decided to post AI-generated images of cats.
--:-- Sean Rameswaram: And every time I post one, I'm going to add an A in the word cat until it's no longer a cat.
--:-- James Vincent: So I was on Twitter one day and somebody actually said to me, as everyone was experimenting with Midjourney, what would it look like? What would happen? Let's see like how many A's you can put into the word cat before Midjourney no longer is giving you a cat.
--:-- Sean Rameswaram: Did you see this?
--:-- James Vincent: Yeah, I did. I did.
--:-- Sean Rameswaram: I did. He got quite far.
--:-- James Vincent: So I did this. And the first cat I got once I upgraded to V4 was just a perfect, beautiful cat.
--:-- Sean Rameswaram: It looked a little bit like my cat, honestly.
--:-- James Vincent: I have like a long-haired, kind of elegant, regal-looking, you know, year-and-a-half-year-old jerk named Gummy Bear who lives in my house.
--:-- Sean Rameswaram: So that was beautiful, regardless of, you know, the depth of the complexity of how it was created.
--:-- James Vincent: And then I put in the second A and immediately I was like oh this is actually probably gonna work because that cat wasn't a normal looking cat anymore it looked I think it looked a little bit more like kind of a steampunk admiral so like immediately second A it was like you don't want just any cat you want a special cat and I was so happy like I was like okay this is going to work Like something interesting is going to happen here.
--:-- Sean Rameswaram: And as I kept adding A's, they kept getting weirder.
--:-- James Vincent: Like there were a couple, like there was one who looked like he was going to sell you weed.
--:-- Sean Rameswaram: Yeah.
--:-- James Vincent: And there was one who looked like he was probably from space.
--:-- Sean Rameswaram: But like a normal, like a cat spaceman, not like an alien cat, but just like a cat who went to space sometimes.
--:-- James Vincent: And then eventually the A's, I think, convinced mid-journey that I wanted cats to scream because there was like a cat.
--:-- Sean Rameswaram: And they began to get kind of more and more angry.
--:-- James Vincent: And also further and further away from a cat until when I hit 10 A's, I finally got my first non-cat.
--:-- Sean Rameswaram: And it was a cat girl.
--:-- James Vincent: It was a human with cat features.
--:-- Sean Rameswaram: I did eventually get one that was a building and definitely not cat-like at all.
--:-- James Vincent: But I never posted that one because it was quite a ways in.
--:-- Sean Rameswaram: And I think, you know, I'm always looking at the kinds of tools that I think that the creators of the future will be using.
--:-- James Vincent: And it seems very hard to imagine a world where this isn't a part of it.
--:-- Sean Rameswaram: Why is it that like eight A's will create a human and not four?
--:-- James Vincent: What is sort of similar to the chat systems you've seen, they have associations that are stored in their memory.
--:-- Sean Rameswaram: What they've done is they've been trained by looking at lots of data on the Internet.
--:-- James Vincent: In the case of the image generation models, there will be lots of pictures of cats on the Internet.
--:-- Sean Rameswaram: Who knew?
--:-- James Vincent: And you know they will be downloaded and they'll be attached to sort of data that describes what is in them so if you are uploading a picture of a cat you probably at some point you know either as a typo or because you're you know having fun with how you're typing it out you've probably added some extra you maybe added some extra a's to that cat however improbable that sounds but there will be a point at which you've added so many a's that no one's ever done that before so the machine no longer recognizes that as an input and and like where in the case of the images where are these images coming from? Are they stealing them from Google image results or what?
--:-- Sean Rameswaram: Each image is being created entirely afresh.
--:-- James Vincent: It's a new image.
--:-- Sean Rameswaram: But what they've done is that they've learned by looking at older images for patterns in them.
--:-- James Vincent: So they take a lot of pictures of cats and they work out what is a sort of common way in drawing or in depicting a cat.
--:-- Sean Rameswaram: But this is sort of been quite a cause of controversy because a lot of the data used to train these image programs is copyrighted it's been put on the web by artists by photographers illustrators and it's been scraped it's been taken without their permission and used by these companies to create these systems and so if the system wants to recreate an individual style you can ask it to do that and it will probably know who that person is and create art that looks like theirs.
--:-- James Vincent: And so the question is, is that copyright infringement? Is there an ethical problem there?
--:-- Sean Rameswaram: You know, is it okay if a human copy is a human? But what if a machine copies a human?
--:-- James Vincent: And what's the answer?
--:-- Sean Rameswaram: It sounds like probably yes, right?
--:-- James Vincent: Well, the legal side of this is very much up in the air.
--:-- Sean Rameswaram: There is one ongoing court case, not about art but about code being scraped from the internet and then used to train an AI where the people are saying that that was done without proper respect to licenses
--:-- James Vincent: But in terms of the art, I've spoken to a lot of lawyers about this in the past.
--:-- Sean Rameswaram: And so the general consensus is that taking the data is legal.
--:-- James Vincent: However, how you then use that data might be illegal.
--:-- Sean Rameswaram: So, for example, you might train a system on someone's artwork, and that is fine.
--:-- James Vincent: You're allowed to do that. That's fair use.
--:-- Sean Rameswaram: However, if you then use that system to create art and then you sell it by saying this is art that looks like this guy's art, then that would be illegal.
--:-- James Vincent: Are there rules or regulations around these kinds of bots yet?
--:-- Sean Rameswaram: Some companies have put rules on them and some companies haven't.
--:-- James Vincent: So OpenAI has quite a restrictive approach to using its AI image program called Dolly.
--:-- Sean Rameswaram: And it makes people pay for it.
--:-- James Vincent: And it's trying to turn it into more of a sort of, you know, a legitimate business.
--:-- Sean Rameswaram: It partners with Microsoft and things like that.
--:-- James Vincent: Other companies have gone, well, the best way for us to do that is to just make this technology available for anyone to use.
--:-- Sean Rameswaram: And that way it will develop faster.
--:-- James Vincent: And their approach is to put these models out without any restrictions at all.
--:-- Sean Rameswaram: We trust people and we trust the community, as opposed to having a centralized, unelected entity controlling the most powerful technology in the world.
--:-- James Vincent: And obviously that comes with these dangers that they will be used for to copy people's artwork or that they will be used for things like generating pornography, for example, which is something that a company like OpenAI doesn't want to be associated with.
--:-- Sean Rameswaram: Are there laws on the books yet?
--:-- James Vincent: Nothing comprehensive.
--:-- Sean Rameswaram: I think this is something where, you know, in classical tech fashion, the technology is moving much faster than the legislation is.
--:-- James Vincent: You know, there are policies in the US, in the EU, aimed at dealing with these sorts of malicious uses of these programs, but they're not very well thought out.
--:-- Sean Rameswaram: They're not necessarily going to come into power very quickly.
--:-- James Vincent: So I think we're going to see problems with this stuff soon.
--:-- Sean Rameswaram: Algorithmic bias like human bias results in unfairness.
--:-- James Vincent: However, algorithms like viruses can spread bias on a massive scale at a rapid pace.
--:-- Sean Rameswaram: If we don't have women and underrepresented people of color at the table, we will bake into those systems bias, particularly artificial intelligence.
--:-- James Vincent: Bias is a huge problem for these systems because they are trained on the internet and the internet is full of a lot of pretty horrible things.
--:-- Sean Rameswaram: There is bias in the text systems and there's bias in the image systems as well.
--:-- James Vincent: So, you know, you might ask a text bot to write a short story about a doctor who is a very nice doctor and lo and behold, the doctor will be white and the doctor will be male because these are the biases that are already present in society.
--:-- Sean Rameswaram: These AI systems, they tend to amplify these biases unless they are corrected.
--:-- James Vincent: There are ways around that. There are sort of tactics and technical things you can do to sort of dampen those biases down.
--:-- Sean Rameswaram: But the number one thing is simply being aware that this is how these systems work and that they have these limitations in them.
--:-- James Vincent: And why is there so much investment or focus on generative AI?
--:-- Sean Rameswaram: Is it because there's some really profitable, practical uses here or was it just a lark?
--:-- James Vincent: I mean, people are betting it'll be very profitable.
--:-- Sean Rameswaram: It's become the buzzword in many ways.
--:-- James Vincent: And if you think about it, you know, tech has sort of been very exhausted in recent years by failures in crypto, by failures in the metaverse.
--:-- Sean Rameswaram: And generative AI actually works.
--:-- James Vincent: You know, these systems actually do sort of what they promise they're going to.
--:-- Sean Rameswaram: And if you look at the potential possibilities you know that they are huge for the creative industries if you're in Hollywood if you're an illustrator then you will have generative AI tools probably become part of your workflow so something that you might do if you're a designer if for example you know you might use a AI program to come up with some ideas so you have to design a new chair it has to be a certain color it has to fit into this environment you might tap those specifications into an AI program, it'll spit out, you know, 20, 30 ideas, and you'll use those as a starting out point.
--:-- James Vincent: I think there's going to be a lot more collaboration with AI in the workplace, that's going to become something that everyone's going to be doing a bit more of.
--:-- Sean Rameswaram: How do we get from from here to there? What are the steps between where we are now and where we'll be when this feels like old hat, perhaps?
--:-- James Vincent: I think they're quite short steps at this point.
--:-- Sean Rameswaram: You know, I had something earlier this week, a friend of mine text me.
--:-- James Vincent: He works at a university and he said, oh, what do you know about chat GPT?
--:-- Sean Rameswaram: I've just been using it.
--:-- James Vincent: And he was like, I just used it to write a macro for Excel, you know, a little bit of code in order to sort a database.
--:-- Sean Rameswaram: And that's something he does a lot as part of his job.
--:-- James Vincent: And he's not a techie guy, but he had used this tool with no prompting to help make his job faster.
--:-- Sean Rameswaram: The tool had written this bit of code for him and then he'd done a bit of debugging it to make it work properly.
--:-- James Vincent: And then he just used it in his job.
--:-- Sean Rameswaram: I think that's the sort of thing that we're going to see becoming more and more commonplace.
--:-- James Vincent: You know, you might not Google something, but you might ask an AI about it instead.
--:-- Sean Rameswaram: James Vincent, catch his work at TheVerge.com.
--:-- James Vincent: Thanks for having me on the show.
--:-- Sean Rameswaram: Our show today was produced by Amanda Llewellyn.
--:-- James Vincent: With an assist from artificial intelligence.
--:-- Sean Rameswaram: Matthew Collette edited Laura Bullard fact-checked.
--:-- James Vincent: And Afim Shapiro engineered.
--:-- Sean Rameswaram: With help from Paul Mounsey.
--:-- James Vincent: And additional help, not from AI.
--:-- Sean Rameswaram: But from Brandon McFarlane.
--:-- James Vincent: From the Vox Media Podcast Network.
--:-- Sean Rameswaram: Who came in with an assist on that Beyonce bit.
--:-- James Vincent: Thanks, Brandon.
--:-- Sean Rameswaram: It's Today Explained.
--:-- James Vincent: All the single patties.
--:-- Sean Rameswaram: All the single patties.
--:-- James Vincent: All the single patties.
--:-- Sean Rameswaram: All the single patties.
--:-- James Vincent: All the single patties.
--:-- Sean Rameswaram: All the single patties.
--:-- James Vincent: Get your cheese out!
--:-- Sean Rameswaram: Support for this show comes from Macy's.
--:-- James Vincent: Fall is right around the corner, and that means layering season is too.
--:-- Sean Rameswaram: Macy's Fall Fashion Guide can help you get a pulse on all the latest trends and must-haves.
--:-- James Vincent: Try a romantic look with a sheer top from INC.
--:-- Sean Rameswaram: Or mix it up with some fun textures, like a faux fur bomber jacket from Karl Lagerfeld Paris.
--:-- James Vincent: No matter your vibe, you can explore the Macy's Fall Fashion Guide and find the perfect look for you.
--:-- Sean Rameswaram: Shop in-store or online at nacy's.com.
