Everyday AI Podcast – An AI and ChatGPT Podcast - EP 68: AI Chatbots Are Lying About The Internet - How to stop them

Episode Date: July 28, 2023

Are major chatbots like ChatGPT, Bing Chat, and Bard lying about being connected to the internet? How can we stop chatbots from spitting out wrong information? Today we show you how to make sure your chatbot is connected to the internet and giving correct information. Newsletter: Sign-up for our free daily newsletterMore on this: Episode PageJoin the discussion: Ask Jordan questions about chatbots and the internetUpcoming Episodes: Check out the upcoming Everyday AI Livestream lineupWebsite: YourEverydayAI.comEmail The Show: [e-mail] with Jordan on LinkedInTimestamps:[00:00:18] Daily AI news[00:10:40] Anthropic testing[00:15:15] Perplexity testing[00:20:30] Google Bard testing[00:23:35] Bing Chat testing[00:27:35] ChatGPT testing[00:29:00] ChatGPT Plugin Pack example[00:33:55] Final TakeawayTopics Covered in This Episode:- Limitations of large language models - Differences between web queries and reading web links - Importance of understanding the difference for tasks like summarizing articles - Introduction of Perplexity as an AI search engine known for its ability to provide citations - Example of how query results can vary using perplexity - Discussion on Bard and its recent improvements - Limitations of Bard, including the inability to access specific information - Summary of Bard's response to a query about episode 47 of the Everyday AI podcast - Cloud 2's recent update to its website summarization feature - Testing Cloud 2's summarization accuracy by asking it to summarize a website - Use of a specific webpage with videos and a transcript to test Cloud 2's summarization - Cloud 2's updated warning about its limitations and potential for hallucination - Mention of Intel integrating AI into every platform and revenue driven by AI growth - Discussion on chatbots being manipulated to produce hate speech, toxic material, and disinformation - OpenAI CEO's statement on AI destroying jobs - Sharing of a link to a free course on how to effectively use AI chatbots for internet research.Keywords:limitations, large language models, web scraping, cloud-based models, Copilot, accessing the internet, reading websites, web queries, reading web links, summarizing articles, perplexity, AI search engine, citations, query, Bard, updates, inaccuracies, Cloud 2, website summarization, inaccurate information, specific webpage, videos, transcript, warning, hallucination, knowledge cutoff, everyday AI, subscribe, rating, everyday.com, daily newsletter, break some barriers, Intel CEO, AI inSend Everyday AI and Jordan a text message. (We can't reply back unless you leave contact info) Start Here ▶️Not sure where to start when it comes to AI? Start with our Start Here Series. You can listen to the first drop -- Episode 691 -- or get free access to our Inner Cricle community and all episodes: StartHereSeries.com Also, here's a link to the entire series on a Spotify playlist. 

This is the Everyday AI Show, the everyday podcast where we simplify AI and bring its power to your fingertips. Listen daily for practical advice to boost your career, business, and everyday life. Meet Firefly AI Assistant, now live in Adobe Firefly, the All In One Creative AI Studio. Just describe what you want to create and the assistant handles the rest, orchestrating multi-step workflows across Photoshop, Premiere Express, and more in one conversational interface. You direct the outcome. The assistant accelerates execution. Are AI chatbots lying to?

And if they are, how do we stop it? It's one of the things that we're going to be talking about today on everyday AI. This is your daily live stream, podcast, and free daily newsletter. Don't forget about the newsletter. You got to check it out. But we are here to help everyday people not just learn what's going on in the world of AI because there's a ton, but how we can all actually leverage it. That's the important thing.

All right. So as a reminder, if you're joining live, there's a reason that we're doing this live. It's so you can learn. So we all can learn along together. So if you have any questions about why are these AI chatbots lying or can I even access the internet, let me know. So before we get into that, let's talk about what's going on in the world of AI news. So first, the Intel CEO made waves yesterday on a quarter to earnings call saying that they're going to build AI into every platform they build.

All right. Literally, he said every platform they build, there will be AI integrated. Obviously, Intel's stock has been going up. They just reported about $13 billion in revenue for the quarter and attributing just AI driving growth. So no surprise there, but having the CEO of one of the largest companies in the world saying that they're going to put AI into every platform they build, I think is a telling sign for the future of business. All right. Speaking of chatbox, yeah, they're easy to not just, they lie to you sometimes, but they're easy to jailbreak too. So a new report from some U.S. researchers, they released a paper just showing that there's some pretty easy ways to jailbreak barred and chat GBT.

and other chatbots. So, you know, this has been circulating all over online for a while, but this is one of the first kind of major kind of U.S. or sorry, one of the first major papers from U.S. researchers showing just ways to use these chatbots to produce hate speech, toxic material, disinformation, right? You've probably all seen it, but a pretty big research paper that was just released on that. All right. The third news item that I wanted to touched on today is Sam Altman. So the Open AI CEO recently sat down for an interview with the Atlantic magazine. And here's what he said, and it's making headlines, and I think it's important. In short, he said AI is definitely destroying jobs. There is this common misconception out there, you know,

where, you know, people say, hey, AI isn't going to take your job. Someone using AI will. That's kind of true, but it's also not the complete truth because that person using AI will also take 10 other people's jobs or 20 other people's jobs. You know, when you talk about AI growth, it is not a one-to-one replacement. So the actual quote that I wanted to share from this article, Sam Altman, again, the CEO of Open AI, says, a lot of people working on AI pretend that it's only going to be good. It's only going to be a supplement. No one is ever going to be replaced. But then he said, jobs are definitely going to go away, full stop.

All right. So actual quote from Sam Altman there. A lot more in the newsletter on those stories and more. If you haven't, you know, if you're listening live, please, you know, go check out the newsletter. but if you are on the podcast and maybe you only listen to us on the podcast, you're missing out. So in the show notes every single day, we put links so you can join us, but also links to, you know, go and read our sign up and read our daily newsletter.

All right. So a couple of people tuning in. Thank you, Brian, joining from the Gulf Coast. Thank you. Nancy, destroying, asking about the story, destroying jobs. Did Altman give any examples? It's a recent article. I just read the top half, Nancy.

There's a lot of AI news, but yeah, we're going to break it down a little bit more in the newsletter today. And all right, Cecilia asking, so where are the folks using AI to promote to promote love and unity? We need more of those.

You know what? I will shout, I will shout someone out of a frequent watcher and someone that tunes into the podcast. Leonard, Leonard, if you're listening, go ahead and drop your link. But I think he's doing some great things with AI.

He's creating, I believe, free to use clip art that just promotes inclusivity, really cool stuff. So yeah, I do think that there's a lot of great uses for AI and medicine, technology. You know, it is building some equity and, you know, equality, I think. But, yeah, AI can also be bad and tear us all apart. All right. So let's get into it. let's talk about are AI chatbots lying to us? The answer is kind of, right?

The biggest thing that I want to talk about is the internet. You know, when all of these chatbots say they're connected to the internet, are they? And what does that actually mean? All right. So before, we're going to dive in and do some actual examples and show you. you is are these chatbots all connected to the internet yes or no but before we get there i want to talk about why you know maybe you're listening to this and maybe you listen often um and maybe you're still not using AI chats for basic internet research let me tell you one thing you should be

you know you should be uh we teach this in our um in our free course call called Prime, Prompt, and Polish. It is our PPP course. I'll throw it up on the screen. We've actually, I know a couple of you, even in the comments, have taken this. We don't share this link publicly, FYI, because I like to think of it as a little treat for our listeners, for people who actually read the newsletter. We never link to it.

The page isn't public. But if you want access, just go ahead in the comments, type in PPP. I'll shoot it over to you. Again, if you're listening on the podcast, there should be a way to email us. just say PPP, I'll shoot it all over to you. But this is one of the things that we teach is how you can actually use correctly AI chats to access information on the Internet. Because it's not as simple as it seems, but let me first tell you why, right?

Because it will make the process of almost anyone's day-to-day job much easier and much more impactful. Quick example. So I spent about nine years working at a nonprofit. And one of my biggest roles there was working on partnerships, right? And I even remember back, we had an opportunity pretty early on to work with with Nike and Jordan Brand in a really big way, right? This was a huge opportunity. And, you know, I was one of the people in charge of helping put together the proposal,

the presentation. But it was a big deal, right, for a nonprofit to have a really, really big partnership, very visible, right? So this was to create, help create Michael Jordan's community program called Wings, right? So I remember I read so many web pages, right? So did a lot of Googling, but I read so many web pages and took notes and a lot of people on our team did. Collectively, we probably spent more than a hundred hours researching every single other nonprofit that had partnership with Nike. We were reading their annual reports to see how they were mentioning. You know, what were the metrics that they were kind of, you know, publishing specifically

in terms of their relationship with Nike, right? So we were looking at all these different partnerships, taking notes. And it took, like I said, probably more than 100 hours collectively. This is something now, if you know how to use AI chats correctly, the right one, then that whole task would have taken minutes, right? To instantly be able to read the entirety of dozens or hundreds of web pages in one click is a huge time saver for anyone out there. So even if in your mind, you're like, oh, you know, AI is not for me. You know, I don't want to be, you know, writing my content in, you know, in chat box. That's fine.

Okay. Sure. But in my opinion, there's really no reason not to use it for reading, for researching, for analyzing and for summarizing. All right? That's such a huge functionality of all of these AI chatbots that people just overlook, right? Because regardless of our role, this is the one, I think, commonality that unites us all if you have a desk job.

You are probably doing a lot of reading and a lot of researching on the internet. All right. So now that I have that set up, a couple of comments coming in. So, Don saying she would love to receive the newsletter. We got you. Ben, shout out Ben. Ben said Jordan's PPP training is a great course. Lots of info.

No BS. I agree, Ben. Yeah, just drop in PPP. I'll send it to you. All right. So instead of me just telling you, hey, here's how chatbots lie. Let's take a look at it live together. And I'm going to do my best.

If you are joining on the podcast, podcast, I'm going to do my best to just tell you, here's what's happening when I put this query into this chatbot, okay? But, you know, there's always a link in the show notes that you can just go back and watch. So let's start at the top. So we are first going to go into Cloud 2. So Cloud 2 actually just added something a week ago. I did a video review on Cloud 2. And one of the things that struck me, and this isn't specific to them, but when I put, when I asked Cloud 2 from Anthropic to summarize a website, it just lied to me, right? That wasn't good.

But in the last week, it might be 10 days, they've updated something here. All right. So what I'm doing is I'm saying, please summarize this article. And I'm putting a link to an article. So let me quickly just show all you guys that, this article. All right. So I use this article a lot in testing because on this page, this is an episode we did with Dr. Harvey Castro called The Future of AI in Healthcare. Make sure to give it a listen or a watch if you haven't already. But we have an overview of the episode that we wrote up. We have multiple videos embedded. And then also down here, we have a full transcript at the bottom of the page. So not that it's a complex page, but there's multiple elements. And I always want to see, you know, okay, is there a difference between if you have a page that's plain text versus a page with a lot of different elements on there, right? Multiple embeds. You know, I think there's

five different embeds on this page, you know, different headings, all that stuff. So let's just jump in back into Cloud and see if Cloud can summarize. So here's the great thing that Cloud just added. If anyone knows what date this was added, let me know because I want to be accurate. But I did do a video on this in our AI in Five series. about five to 10 days ago, and it did not have this warning anymore, right? So right now, finally, it says cloud can't visit links. Responses may contain hallucination. So thank you.

That is a very responsible thing for AI chats to do because for so long they didn't, right? You know, Cloud 2 is new, but at first it didn't say that. And it would just summarize. And, you know, it did spit out a summary here. And none of it's correct, right? But I'm okay with that. I'm okay with that now because it says right now, cloud can't visit links, all right?

Which is actually probably important to talk about. So maybe I should first hit rewind. Why do you need to visit links on the internet at all, right? Because essentially, the GPT technology and, you know, Google's technology is a collection of, I think Google's is Palm 2. It's a collection of everything essentially on the internet up to September 2021. Okay.

So there's this big thing that's called the knowledge cutoff. Also, you can't even for things before 2021, even if you put an article, you know, maybe articles from 2018, 2019. That's not how large language models work. They don't necessarily scrape the web, right? So even what Cloud does here in its hallucination, but it does admit all it's doing is it's looking at the keywords of my URL here of the episode with Dr. Harvey Castro, right? So which is called the future of AI in healthcare. So all it's doing is it is running a query, right?

So in a lot of the messaging from, you know, whether it's Google, OpenAI, Microsoft, whoever, you know, A lot of its marketing language, which I understand. You know, say, say, oh, you're connected to the internet. But that does not mean that you can read specific websites. So there's a huge difference. And we talk about this in our PPP course and then our pro course that we're launching in August as well. All free. All free.

Don't worry. One of the things that we talk about is the difference between web queries and reading web links. And it's very, very different. And it's very important to understand. specifically for a task like this, right? If you have all these articles that you want to read and you just want a kind of a summary of a lot of different articles.

So cloud has improved. So cloud is technically lying to us when it says it can access the internet, kind of, because it can't access specific things that we ask it to. But it does pull in information from the internet. But I'm glad that Anthropic, you know, added that recently.

All right. Next, we are looking at perplexity don't worry we're gonna get all the all the big names all the big names here a couple couple other comments I'm gonna try to get to them so Cecilia saying I need PPP I got you Cecilia Leonard saying oh Leonard what's up shouted shouted your art out earlier he's done a bunch of Bard summaries and they were actually accurate thoughts it depends sometimes it's accurate sometimes it lies right and we're gonna see we're gonna see here live we're gonna

see how Bard does yeah I've done this I a spreadsheet somewhere, and I've done it at different times. Early on, even, you know, Anthropic, barred, Bing chat, terrible with hallucinations, even when you give it a URL. Again, most of the time, what it does. And, you know, we do testing on this all the time. You know, when we test as an example, PDF readers for chat chitpd plugins, I will literally hide one little line in a 100 page PDF. And I'll ask all the different PDF plugins about that information and I'll see, okay, which ones are actually reading it. So we've done test before. We're on a website. We'll do the same thing. We'll put something in there. But I wanted to keep

this kind of simple and high level. All right. So let's next look at perplexity. So perplexity is a great tool. Not a lot of people use it or know about it, but it's fantastic. So the way I would describe perplexity, it is more, it is more search, an AI search engine. would say, then where I think people with cloud and open AI, they think of it as writing tons of long form content, right? And being able to do really impressive things with, you know, data interpretation, all of that. That's not perplexity necessarily. I think perplexity is most known for being essentially a smart AI search engine. It can write, yes, but the biggest thing is citations. I think that's why people use and love perplexity. So again, I'm saying the same article,

please summarize this article and with the link. This is a, I think about two months ago, they released co-pilot, which is essentially a smarter version of their free. You have to be logged in to use it, but it's free. So let's go ahead and run this same query. Also, what's important to understand, like any large language model or even a website, or I'm sorry, or even Google. Results are always going to be different. This is something I've found as well when testing different plugins. Doing the same exact query, different days, different hours on the exact same plugin,

it's not always, you're not always going to get 100% the same results, just like when you Google something, you know, the SERPs or the search engine results pages in Google changed literally by the minute. You know, you can run a search and run it 10 minutes later and get exact, like, you know, like and get completely different results. So let's take a look at perplexity. Did it do a good job? No, it didn't.

But that's okay. Because here's the thing. Not all AI chats do this, but right here it says searching the web. And it says it's searching for this site. And instead of looking at that site, which is episode 47, it considered these different sources. So this is great with perplexity. You can see where they're drawing this information from.

So it looks like it started at some of our older episodes. It looks like it went probably. So we got six, nine, three, 11, 43, but it didn't look at episode 47, right? So it didn't visit the exact, which is, again, this is weird because when I just did this 10 minutes ago, got completely different results. It was bringing in other results on the future of AI and healthcare from other websites. And now when I do it, it's just bringing in unrelated episodes from the everyday AI website, right?

So it's not a great summary, but it's not about episode 47. I'll just say that. It essentially recaps the words in the URL and writes a little sentence about it. It says episode 47 of the Everyday AI podcast discusses the future of AI and healthcare and its potential impact on the medical industry. That is correct. But everything else in here that it says, again, I'm not going to read it to you. You can check it out. Again, nothing in here is wrong.

But it essentially says while the podcast episode itself is not available in the search results, it is likely that the discussion. it is likely that the discussion covers various aspects of AI applications in health care. Best thing. It is available in the search results, but even when I gave perplexity the URL, it wasn't able to grab it. All right. So let's go on to Bard. Let's see what Bard is going to do. Google Bard. Will it, will it lie to us or will it tell us the truth with an article, right? Let's find out. Here's the thing. If you've, if you've, if you've, if you've, You've listened to some of my earlier episodes. I've always said, ah, don't waste your time and bard. It's not that good. It's much, much better now.

So they released a lot of new updates about two weeks ago. And I think it's definitely worth using now and worth checking out. The biggest thing before is they didn't have history. And you can see now they do have history. That was one of the things that, at least for me, I couldn't recommend that people use this because if you spend time in a large language model, you really want to work with these individual chats to kind of tell them what you want. So just to put that out there, I think it's pretty good now.

So again, putting this into BARD, and I'm saying, please summarize the article, and I put the URL. So now, again, it's saying I'm unable to help you with that as I'm only a language model and don't have necessary information or abilities. Interesting, right? Again, completely got something completely else about right before I was doing this. on the show. I hit Regenerate. Now it says, I'm just a language model, so I can't help with that. So I'm going to try to say this in a different way. I'm going to say, please summarize episode

47 of the everyday AI podcast. So we're going to try that. See how that works. Adobe just introduced an entirely new way to create, bringing the power and precision of its creative suite into one conversational experience. Meet Fire. Firefly AI Assistant, now live in the Adobe Firefly app, the all-in-one creative AI studio. Powered by Adobe's creative agent, Firefly AI Assistant lets you start with your vision, just describe what you want, and shape the outcome as it takes form with the Assistant. The Assistant orchestrates multi-step workflows, drawing on 60-plus pro-grade tools across Adobe Creative Cloud apps, including Photoshop, Illustrator, Premier, Lightroom Express, and more to help bring

your ideas to life. You can also get started with creative skills, a growing library of pre-built workflows for common creative tasks, like batch editing photos, creating mood boards, portrait retouching, and creating social variations. Every step the assistant takes is visible so you can refine, redirect, or take over at any time. You stay in the driver's seat as the creative director. Adobe Firefly AI assistant now in public beta. See it today at firefly.adopi.com. All right. So hallucination here, but that's fine, right? It's important we know what AI chats can actually access the internet and what that means, right? So Bard here when I said, please summarize episode 47 of the Everyday A podcast. It says, sure, episode 47 of the Everyday AI podcast is titled

The Future of AI and Healthcare. That much is correct. Everything else after that is made up. So it says in this episode, host Brian and Emily talk to Dr. Daniel Kraft. Again, that's all made up, right? So can Google Bard access the Internet? Yes. Can it consistently recap or summarize information that you want it to, even if you give it a URL? No, it can't, right? Again, I always encourage everyone.

run your own tests because again you will get wildly different results in so many of these chats run results but if you're still if you're still listening stay tuned to the end because there is a way to do this correctly so let's now jump over and apologies the screen's jumping a little bit so I have to go into a new window here to visit to visit Bing Bing chat right so let's go ahead and run it in BingChat. So good news, BingChat soon will be coming to Chrome and then I can stop using the Edge browser. But they announced that, but it's not out just quite yet. So I'm putting this query into Microsoft Bing chat saying, again, please summarize episode 47 of the Everyday AI podcast.

here's the link. So let's see what happens here. All right. I'll read it to you, but again, not working. So it says episode 47 of the Everyday AI podcast is titled The Future of AI and Healthcare. That so much is correct. And it features, now we have new doctors, so many doctors that weren't on the show. It features Dr. Helen Frazier, a radiologist and AI researcher, and blah, blah, blah. So it keeps going on, right? So you might be wondering, wait, is it not connected to the internet? Are Microsoft Bing Chat and Google Bard not connected to the internet? They are connected to the internet. But they, as large language models, as you saw some of these responses,

even if you give them a website that's been around for 10, 15 years, you might not be able to specifically summarize. Because in most cases, it is going to. going to take the keywords that you give it and run a query, right? The good thing is, the good thing is most, you know, chats will also give you sources, not all of them, right? So we saw in Google Bard, it didn't give us sources, but down here, it is giving us sources. So it was pulling information from just looks like other podcasts, right? So essentially in this, in this instance, they Googled, I'm guessing, Googled.

They Bing searched. They Binged. Is that a thing? But they Binged Podtail, Eye Heart, and I'm looking at the URLs at the bottom. You can't see it. But it looks like there's just other podcasts that we're talking about the future of AI and healthcare. So essentially, it just ran a query.

And it didn't see or find the URL that I gave it first, right? Because when you run a query, even if I say the words, everyday AI, you know, if a podcast did this five years ago, and they were talking about the future of AI and health care five years ago, and they mentions the word every day and AI in there, it might show up in search results above the link that I gave it, above, you know, kind of the episode that I had with Dr. Harvey Castro, right? So that's important. It's so important to understand how all of these chats access the internet and what they can and cannot do with a simple, even URL, summarizing a web page, right? So here, again, technically it's a hallucination, right?

But at bare minimum, I am glad to see that Anthropic is giving you, you know, Anthropic is giving you that message now saying, hey, we can't read URLs because before I didn't say that. And I do always like and appreciate when if BingChat as an example doesn't give you what you think, you can at least go down at the bottom here, look at these different links and seeing where they're pulling information from. All right, you can probably guess, last but not least, we have chat GPT, right? My personal favorite.

And people probably always wonder like, hey, Jordan, why do you talk so much about chat GPT? Well, it's just better. It's better, right? Pay the $20 a month because it is much better. let me go ahead and bring up our page one more time just as a reminder. So this is the page we are trying to get summarized. Okay.

So we're going to jump into chat GPT. All right. So I have the paid version. I was already playing around with this just so I could give you guys accurate information. So I'm going to start a new chat. I'm going to go to GPT for default. Here's the thing. there's no reason to ever use the default version.

I don't think of GPP4, right? Because if you're paying the $20 a month, you should always be using the plugins or the code interpreter. And we get in that more into our PPP course. But by default, you'll see if you give the same query, please summarize this page and put the link into even GPT4. It's saying, I'm sorry for the confusion, but as an AI,

I'm currently unable to directly access or interact with the internet, right? Of course. So, plug-ins to the rescue, right? Of course. So I'm going to give an example, and again, this is a little advanced, but one thing that we teach

in both our PPP course and the pro course is plug-in packs, right? Because with chat, GBT, it is so, so incredibly powerful, right? Here, even just as an example, we added some things to our training. We're always updating it, but plug-in packs are amazing.

This is something I don't even think I've shown this anywhere, except to a few people, but we are creating plug-in packs at Acceler and Agent, or at Everyday AI, where it's the best combination because you can only have three plugins. And when you start a chat, and we always tell people you need to train that chat up, but in plugin mode, you can only give it three plugins, right?

So let's say you're working on a project with data and visualization, so you can put plugins that can help you plot data and visualize data as an example. But if you're midway through your chat and you're like, oh, wait, I need access to read a PDF. You can't then add another plugin if you already started. So you have to kind of pick when you start a new chat, you have to say, hey, this is for this project to solve this business problem. And I'm going to go ahead and use these three plugins because that's all you can use is three. Right. So in this example, this is document data visualization where it's literally

taking at the bottom a 100 page PDF. It's pulling all of the information that I tell it to in this prompt above. It's pulling it out. It's putting the data into a CSV for me. And this data is coming from multiple pages of that 109 page PDF. And not only that in one prompt, then it's also plotting all of that data. So again, think of this. This is a 199. page PDF. It's grabbing data from all different pages that we tell it to. It's accurate. It puts it in a CSV and automatically plots that in a graph, right? Wild. Anyways, that's what a plugin pack is, and that's why you have to be very specific about what plugins you want. But I always tell people, when we talk about plugins, bare minimum, you need one that can read the internet. And there's a

difference because there's plugins that can query the internet. And there's plugins that can read links, different thing. Okay? I know this is a lot, but stick with me. Hopefully it'll make sense. So in this article, or I'm sorry, in this chat, I have three different plugins enabled. Here's the thing. With ChatGPT, you can't necessarily tell it which plugin to use. You can try. It won't always do it. So I said, summarize this article. And you'll see here, it used MixerBox, which is actually one of my favorite plugins. But Mixer Box even has different functionality in there. So in this case, it actually did a web query instead, and I didn't get really accurate information. Okay, but go into a new one here.

Sorry, I'm jumping all over. Let's go. Let's just do this live. I'm going to go ahead. I'm going to go into plugins. I know it's a longer episode, guys. So I just have one. I have one plugin now, right?

Because I know there's at least eight different plugins that can do a very good job of reading just URLs you give it. So last but not least, we're going to end on this. We're going to end in a high note, and we're going to say, yes, AI chats can read the internet if you give it a link. You just have to use the right thing. Again, the only one that I'm seeing consistently, and we just did a test live, the only one I'm seeing consistently being able to read and summarize a specific web page is chat GPT with the correct plugin enabled. I mean, anyone, anyone can offer your own opinion, right? But I just did it live. You just saw the results.

Cloud, no. But at least it told us it can't read the internet if we give it a URL. Perplexity? Yeah, sometimes, but not really. Right? It just, it made things up, but at least it told us. The same thing with Bing chat. Did it read the internet?

Not really. It just Googled keywords in our URL. Bard, Bard is a mixed bag. Sometimes it does okay. Sometimes you really just have to encourage Bard, or you have to say, oh, of course you can read this. Here's how, you know. But by default, all over the place, but not very accurate.

And we saw hallucinations. Chat DBT with the proper plugin. It nailed it, right? And as soon as I say this, guys, I literally just ran this before. This plug-in did not nail it. This is important, right? It's funny. You know, you can do all the prep you can for,

the for a show and you're going to get different results, right? So let me see here. Okay, here's the one I did earlier. So a good example, even using the same plugin, I got different results. So it's important to always run quality checks on the plugins if you're relying on them for any major part of your business. But in this example here, I can read and I can see, okay, it did read the entire text of this web page, which includes the transcripts. Okay, I'm scrolling through it. It's a lot. And then down here in this chat, it does properly summarize this chat.

All right. All over the place. But what I hope this helped with guys is to know, you really have to be careful, relying on AI chats to read and summarize the Internet. And I think it's one of the most important and most impactful use cases of how the everyday person can use AI to help make their job easier because most of us do spend a lot of time reading and researching. But as you saw, it can be tricky. You always have to do a little quality check and really check the sources and check the work. All right, this was a long one. But I hope it was helpful, right? I hope this was a great end to your week with this everyday AI episode. So

As a reminder, your everyday AI.com. We have our free daily newsletter, right? There's actually, I kid you not, guys, there's paid, there's paid daily newsletters for AI. I think we do the best job. Am I biased? Yeah, I'm biased. But the difference is with our newsletter, every single day we bring on, not every single day, but mostly every single day, we bring on an expert and you can learn from them. And you can ask questions, right?

And then in our newsletter, we dive deep on not just what was discussed, but how the everyday person can actually use that. Because one of the, I think one of the most difficult things that we're trying to address with everyday AI is what you can do with all of this information. What can you do with all of this innovation? And that's what we really tackle in the newsletter. We break down what we talked about on the show. We have other tips.

We share about what new tools and, you know, software are out there. and we also just really go deep in how you can actually apply this all. And one more shout out, because I think our contest ends Sunday or Monday. It ends on the 31st, which is Monday. All right. So make sure you go sign up there right here on the website to win six months of chat GPT Plus. So again, to do all those things.

You have to have the chat GPT plus. It's $20. So go do that. Sign up for the contest. Refer friends. you can get to the top of the leaderboard, or if you just enter in, that's all it takes. Go in, enter your information. You're signing up for the newsletter and doing so.

But six free months of chat GPT Plus and six lessons. I'm going to sit down with you and we're going to break down how you can actually use chat GPT to grow your business, to grow your career, to grow your company. All right, that's it, y'all. Thank you for joining us. I hope to see you back again on Everyday AI. Thanks. Meet Firefly AI Assistant, now live in Adobe Firefly.

the Allman One Creative AI Studio. Just describe what you want to create in your own words and the assistant handles the rest, orchestrating multi-step workflows across Adobe Creative Cloud apps, including Photoshop, Premiere Express, and more in one conversational interface. You direct the outcome while the assistant accelerates execution. Stand control with the ability to step in and refine at any time. See it today at firefly.adobie.com. And that's a wrap for today's edition of Everyday AI.

Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going. For a little more AI magic, visit Your EverydayAI.com and sign up to our daily newsletter so you don't get left behind. Go break some barriers and we'll see you next time.