AI's big year and the tools changing our lives

Transcript

CHATGPT: "Ever chatted with a robot who's better at conversation than your last date?"

NIALA: As AI keeps getting better, how do we keep up? AI's big year and how it's already changing our day to day. I'm Niala Boodhoo. And from Axios, this is One Big Thing. Happy Thanksgiving, I hope you're listening to this while getting ready for a delicious meal with family and friends, or finding rest and joy some other way today. Or if you're in holiday traffic, well we are here for you in that case, too.

This week there's been a lot of news about leadership at Open AI, just one tech company that's leading the way in this space. But even amidst that chaos, the technology keeps marching ahead.

So today for the holiday we are bringing you a special episode featuring one of our OWN leaders, Axios chief technology correspondent and co-author of the AI+ newsletter, Ina Fried. She's been covering tech for twenty years and is keeping close tabs as AI evolves. She's got the latest and will walk us through a handful of AI tools already available today that could make our lives easier, but they have some downsides.

So let's catch you up quick on how AI has exploded in the last year.

INA: So in the last year alone pixelated photos with seven-fingered hands are now photorealistic images you can control with precision, for example. We have chatbots that once knew nothing beyond mid-2021 are augmenting their web-based training with input from voice, images and current events.

NIALA: To keep up with all this change…Ina has been testing out and writing about AI tools as they emerge and improve…from one that lets you create your own personal deepfake…to an AI assistant that lets you attend multiple meetings at once. What you heard at the top of the show?—"Ever chatted with a robot who's better at conversation than your last date?"—that was ChatGPT's response to Ina's question "what would be a funny intro to a video about ChatGPT?" I asked the real-life Ina what we need to know about some of the MANY tools out there right now…and why they matter. Hey Ina.

INA: Hey Niala.

NIALA: So AI tools are available to the masses. I mean, we can all access these kind of things. And what we're talking about today are the things we can all use and how that's affecting our day to day lives. So with that, when you look back at the last year, what's the single biggest leap in AI technology for everyone that you think we've seen?

INA: I think the big thing is we've gone from one sort of abstract engine, chat GPT, to many engines, but also training on specific data. I'm still a big believer that the real revolution comes not with these tools on their own, but when you take this powerful tool and combine it with specific sets of data.

NIALA: ChatGPT is arguably the tool that cracked open AI for the masses. It's probably what everybody hears about the most as well. What does it do now, and what's it like to test it out?

INA: So, yeah, ChatGPT, not even a year out there, and it's added a couple really significant capabilities that I think are meaningful in the moment, but exceptionally important when you think of where they're going to take us. So, now in addition to being able to type text into ChatGPT and ask a question, you can send a picture, so you can have a picture of your refrigerator and say what can I make for dinner or a picture of a math problem and it can answer it directly from that. You can also converse with ChatGPT.

Why I think that's super important is it opens up a world where you have these engines that not only understand what they were trained on, but can make sense of the real world. And I think that combination opens up a lot of possibilities, some of them rather scary, but I think it won't be that long before we have a robot out there with chat GPT as its brain. And again, I'm not convinced that's a good thing, but I do think it's a powerful thing that's coming.

NIALA: So, how well does chat GPT do those things? For example, did it tell you what to cook?

INA: So, it's medium. So my refrigerator is really packed full and you can hardly see what's there. And, no surprise, ChatGPT can't see past the leftovers either. So, uh, I think it's impressive, but not yet foolproof.

NIALA: Ina, one big development you've been writing about is CUSTOM GPTs and mixing generative AI with specific business data. Is that just like a custom chatbots or is it something different?

INA: It is, but the real power here is we're marrying all the ability of GPTs to reason. I hesitate to use the word, you know, smarts because I don't think they're actual smarts, but combining that with specific data. So whether that's you creating a custom one or specific industry ones, I think that's the real power we're going to see over the next year or two is when people take this technology, but they really ground it in something. So what OpenAI announced at its Developer Conference was the ability for anyone, and I created a couple of these, to create a custom version of ChatGBT that has extensive knowledge about a particular document or set of documents. So I created one to help prepare for my Olympics coverage next year.

NIALA: So…different from OpenAI, we're seeing custom uses of this kind of tech in the medical realm, too – like, how could this help doctors?

INA: So, for example, I broke the story about Elsevier, the big medical publisher which has access to all these big medical journals, partnering with an AI startup so that doctors will be able to ask questions of not just the generic web, I don't want my doctor doing that, I don't want to do that - but asking questions of the latest medical journals, and so they can either look for drug interactions or symptoms, that sort of thing.

And again, the difference here is, it's not that ChatGPT or these generic tools aren't incredibly powerful, but they do hallucinate, which means that they are based on the whole web, and some of the web is crap, you know. All that ivermectin stuff is still on the web. You don't want your doctor looking at that. Whereas, I think combining this type of technology with all the medical research... Don't forget, doctors go to medical school once, and that could have been five years ago, it could have been 25 years ago. And the rest of their knowledge is what they can cobble together in their spare time. And they can't read every medical journal. AI can.

NIALA: Let's talk about another tool, and this is one that lets you create your own deep fake. That's an AI-generated video double capable of reciting virtually anything that you type into a text field. First, why, what's the good use of this technology? Cause it's hard for me to see anything positive coming from this.

INA: There are some uses, uh, they may not be the most obvious things. So, things like corporate trainings, instead of an actor or an executive at a company having to voice the whole thing. If you create a video double, an AI double of them, you can just type in the words. You can do some custom marketing, although, you know, once that capability is widely understood, I don't think it's all that impressive.

You know, it's one thing if some famous person records a video for you, it's another thing if their avatar records a video that's been typed in. Not as impressive, not as exciting. Uh, but I do think, I do think this has some uses.

I think it will have a lot of uses in terms of mocking up things for, um, all sorts of creative people. You can, you know, basically storyboard out a video, whether that's a movie for Hollywood or an internal corporate training using AI to be the actor, at least for the rough cut.

NIALA: So I have a question. How believable you thought that Ina Double was when you tried it out? I will say, when I watched the video, I just thought, that's weird. It doesn't look like, I don't ever think of Ina as someone who would read from a script. And it seems like she's reading from a script, but that was the only thing I noticed.

INA DEEPFAKE: I'm Ina Fried, Chief Technology Correspondent at Axios. For my latest prompt review, I am taking a look at HeyGen, a new video service that uses AI to help create automated videos.

INA: Yeah, I mean, it's pretty good. And when I look at these technologies today, I assume this is a starting point. I assume most of these tools and everyone I talk to in the industry says it keeps getting much better from here. So I think it's pretty darn good today. Um, yeah, it sounds a little flat. The lips aren't perfect, but it's pretty good today. And I think it will be virtually indetectable from the naked eye and ear tomorrow.

NIALA: So then the obvious question is, how on earth do we keep track of what's real with tools like this out there?

INA: I think we're going to need what's called Content Authenticity. Um, so Adobe has this Content Authenticity initiative. There's other projects too that are trying to authenticate how a video or an image was created. And you can track the provenance from when it was captured or recorded through to where it is and see what edits have been made.

The challenge is, you know, good actors are willing to do that, bad actors are not going to label that they're generated. But if you have enough content that is authenticated, hopefully we get to a place where, as a society, we're even more skeptical of something that's not authenticated. But in order for that reality to happen, we need these authentication tools to not only exist, but be incredibly widely used.

NIALA: Quick break and we'll have more with Ina Fried on some of the AI tools out there right now. Back in a moment.

Welcome back to 1 big thing, from Axios, I'm Niala Boodhoo. We're digging into the explosion of AI tools available now with Axios chief technology correspondent Ina Fried, who has been trying many of these tools out herself…Okay, so Ina I'm really interested in this next one…you tried a tool called Lipdub that uses AI for video translations?

INA: Yeah, so this technology, I mean, they're one, but this technology is here, and it's going to be pervasive, which lets you speak what you want to say in your own voice in any number of languages, um, and that has relevance and importance for people who are content creators, like individual influencers now, can get their word out in other languages. So it opens up new revenue, it gives them more reach, and it's also going to have a profound impact on Hollywood, where there's been people that have made their whole career, for example, by being the Arnold Schwarzenegger of France.

DANIEL BERETTA: "Je reviendrais."

INA: This technology would allow Arnold Schwarzenegger to be the Arnold Schwarzenegger of France, because you can speak in your own voice, in your own intonation, speaking a language you don't actually speak.

NIALA: And it seems like translation is such a rich area for using AI.

INA: It is. I mean, it's one of the things I talk about, a few things that AI is really good at. And translation is one of them. Sometimes translation means, you know, shifting from one genre to another. You can ask ChatGPT to put it in iambic pentameter. But sometimes translation literally means language translation, and it's really good at that.

NIALA: You also tested out a tool that makes it seemingly possible for us to be in two places at once? In fact, your Otter pilot AI assistant joined this call when we first started!

INA: Yeah, so I love this technology. I mean, you know, there's, there's some issues it raises. I certainly try and let everyone know, um, that I am recording. That's required by law in California. And, and I think it's just good practice wherever you are. Um, but what it does is it records a meeting. Um, it records an in-person meeting if you want, but also it can join a Google Meet or a Zoom meeting or a Microsoft Teams meeting, and it not only records it and transcribes it, which by itself is pretty impressive, but it can also be queried, so it has its own little chat GPT-type engine built in, an AI assistant, where you can ask, what were the main points of the meeting? Can you summarize it? And a lot of different tools are being added to these various meetings. Otter is nice because it works across the different major ones, but it's really a game changer. And one of the cool features is, it does let you be in two places at once. So if I have two meetings on my calendar, say, You and I are supposed to chat for the podcast, well I have to do that in person, but what if there's a corporate training at the same time?

Then my Otterpilot can go, take notes, transcribe it, and I can either watch the full thing afterwards, listen to it, or I can just say, Hey Otter, what did I miss? And it'll summarize it.

NIALA: Ina, that seems to be an incredibly practical application that I think everyone can relate to. And when we think about the ways AI tools are making their way into our lives, how much of it is going to be things like that that are tangible that we see?

INA: I think it's going to be a mix. I think a lot about this technology will be applied behind the scenes, so you will be chatting with AI. Maybe it'll tell you. I think in general the best practice is that AI should disclose when you're chatting with AI and not a human. But I think we'll see it in a lot of places, you know, when I think of customer support, you know, that's an area that's ripe for this. Um, just querying, you know, shopping. You're gonna, anyone that provides goods is gonna want to let you have an AI to answer questions. So, both Amazon and Walmart are doing things like this.

NIALA: What exactly are Amazon and Walmart doing?

INA: So they're looking at how can AI help summarize reviews that are out there, or product descriptions. That's something that Amazon's looking at. Walmart is looking at, how do they bring conversational AI so you can ask questions of products, uh, which phone is right for a 10-year-old, or I want to decorate this room for the holidays, give me some ideas. That's an example of where it's low risk. It's handling a particular task, it's doing it in a way, and it still needs to warn you that it's AI, and they do, and that it, you know, might do something crazy, because it might, but I think that's an example of where it's not trying to change the world, but it is trying to say, how can we make this experience a little bit better using where today's AI is good.

NIALA: Are there other ways that AI is affecting our daily experience as humans that we might not be aware of?

INA: I think it's coming to all manner of creative expression. So that's things like photography. We're seeing tools that let you basically create images that are better than what's reality. So the question is, how do we feel about that? A purist might say, well, it's not really photography if you're moving stuff around afterwards.

And you might be like, yeah, but I just want a picture where all three of us are looking at the camera and smiling. So if it's doing magic behind the scenes, all the better. So I think we're going to have a lot of new capabilities, a lot of new ethical questions. I think we are going to see AI do parts of lots of jobs.

I think what we're seeing right now is the mistake that some companies are making by trying to have AI do too much, too soon. I think it will be capable of a lot, but I think any time you try and have it do a whole job, it tends not to do as well as when you have it do certain tasks that it's well-suited of.

NIALA: So who's doing too much too soon?

INA: I think we've seen it in journalism. I think we've seen media outlets that try and have ChatGPT or AI like ChatGPT write entire articles and, you know, AI will make stuff up and it's hard even for human editors to spot all the mistakes. That, I think, is an example of doing too much too soon. Uh, we saw a lawyer reply to a court brief with ChatGPT-generated response and it wasn't accurate. It made stuff up. Whereas I do think long term, AI is going to dramatically reshape the legal industry and AI will help lawyers in immense ways. But if you turn over 100 percent of a task to AI right now, you're probably making a mistake.

NIALA: I feel like this is a good time to point out to our audience that Axios does not do this with our journalism. We have made a commitment about the fact that humans will be doing the journalism and writing the stories.

INA: Definitely, and I think that's the right decision for right now. I think companies like ours, like in every industry, I think it's right to say, what can the technology help with? I think it's probably a mistake to give it carte blanche, and whatever you do, I think it's really important to be transparent with your audience, whoever that is.

NIALA: You know, you mentioned 10 year olds and choosing phones. You're a parent. How do you talk about AI with your own child, if you don't mind sharing that with us?

INA: No, definitely one of the reviews I did was this project, uh, called Kid Jenny that's, you know, a small project, but the reason I picked it was it did raise this question of how do we as parents interact with AI with our kids? When do we let them try it out? And so I thought it was important, even with my 10 year old, to say here's what this technology is and this tool that I reviewed lets you So, uh, Harvey had it draw, um, I believe it was Steph Curry playing the trumpet.

And it did a pretty good job, and it also writes bedtime stories. From a very small prompt, it writes a whole bedtime story. And I, I thought that was important because we can't live... burying our heads. This technology is here. It doesn't mean we should use it for everything, but we need to understand it as parents, and our kids need to understand it so they know what it can and can't do.

Um, I'm not going to let him do all his homework using ChatGPT, but it doesn't mean I'm going to ban him from using it. I think it's important that he know how to use these tools.

NIALA: OK and Ina I have to ask – given the day we're releasing this episode, how might AI be showing up in people's Thanksgivings this year?

INA: Oh, you know, I think all kinds of things. I think it is one of these technologies that is so accessible. And I think the thing that I want to think about this Thanksgiving is how can this technology help people like my parents that are not very tech-savvy? I think there's a chance this could really break through because suddenly you're able to just describe what it is you want versus having to learn the language of the device.

And if we can get there, and I feel like we're so close, that could be really profound for a generation. And I'm looking forward to the day and looking towards the day, maybe not forward, where it's my kid that's the tech expert and I'm the one who struggles to use the newfangled devices. So this could help me too.

NIALA: Axios's chief technology correspondent, Ina Freed. Thanks, Ina. Happy Thanksgiving.

INA: Happy Thanksgiving, Niala.

NIALA: And you can find videos of Ina's reviews of AI tools, including ChatGPT and the HeyGen deepfake creator, in our show notes.

That's all for this week's edition of 1 Big Thing. You can always send feedback by texting me at 202 918 4893, or emailing podcasts @ axios.com.

I'm especially grateful for the 1 Big Thing team, which includes Supervising Producer Alexandra Botti and Sound Engineer Alex Suigura, who also composed our theme music. Aja Whitaker-Moore is Axios' Executive Editor, and Sara Keuhalani Goo is Axios' Editor in Chief.

I'm Niala Boodhoo. Thanks to all of you who are doing tech support for your family this holiday, in realtime, without AI… thanks for listening, stay safe, and we'll see you back here next Thursday.