Everyday AI Podcast – An AI and ChatGPT Podcast - EP 350: The Correct Ways To Give ChatGPT Access to the Internet

Episode Date: September 4, 2024

Win a free year of ChatGPT or other prizes! Find out how.You're prolly doing this wrong. But when using ChatGPT and other LLMs, you've gotta understand how and when the models connect to the internet. You think you're getting up-to-date and accurate information. But you're probably not. We'll break it down and show you how to do it the right way.Newsletter: Sign up for our free daily newsletterMore on this Episode: Episode PageJoin the discussion: Ask Jordan questions on ChatGPTUpcoming Episodes: Check out the upcoming Everyday AI Livestream lineupWebsite: YourEverydayAI.comEmail The Show: [e-mail] with Jordan on LinkedInTopics Covered in This Episode:1. Understanding ChatGPT's Internet Interfacing2. Correct Usage of ChatGPT3. Problems with Outdated Data in ChatGPT4. Chain of Thought Prompting With ChatGPTTimestamps:02:15 Daily AI news07:10 Getting access to the web in LLMs09:25 Use Bing for current, accurate information. RefineQ process.13:01 Large language models may be overly verbose.18:55 Bing integration into conversation mode, quick feature.19:44 Need to call browsing with Bing carefully.24:49 Content ranking tricks reveal dirty search secrets.29:18 Research recent LLM developments from August-September 2024.31:32 Wide variety of sources yields better results.34:29 Trend of 2024: Retrieval Augmented Generation, data connection.37:55 Use Bing in chat GPT for URLs.41:09 Stay vigilant, monitor GPT-3, control browsing.Keywords:ChatGPT, WebPilot, Internet-connected GPT, adapting GPT behavior, math problem solving, "browse with Bing", information retrieval, outdated information, incorrect information, relevant information, context window, AI University, thanks a million giveaway, Jordan Wilson, Everyday AI, AI news, Salesforce, Anthropic, Department of Justice, NVIDIA, OpenAI, prompt engineering, YouTube, Bing, old knowledge, large language models, training data, Microsoft, chain of thought prompting, patience in AISend 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. When you're using chat GPT, you probably think that you're connected to the internet.

You see that little icon come up. So you think that everything that you're getting from the outputs is going to be up to date and correct and accurate. And you're blindly trusting it. But you're wrong. Because unless you actually know what you're doing, there's a high likelihood that even when you're getting that little icon in chat GPT that assures you you're browsing with Bing, you're actually really not, or at least not doing it in the way that you think. But that's okay.

We're going to go over that today and help you fix it as we go over the correct ways to give chat GPT access to the internet. All right, I'm excited for this one. What's going on, y'all? My name is Jordan Wilson and welcome to Everyday AI. This is for you. It's a daily live stream podcast and free daily news. that are helping everyday people like you and me, not just learn generative AI, but how we can

leverage it to grow our companies and to grow our careers. And if I'm being honest, chat GBT is probably a tool that almost all of us use if you're listening to this podcast. And this is one thing that I've seen consistently people get wrong, even people who use chat GBT for hours a day. So we're going to fix this. But before we do, got to go ahead and call this out. If you haven't already, make sure to go to your everyday AI.com. Sign up for our free newsletter. Also, hey, on the website, it's like a free generative AI university. There's hundreds of episodes, blog posts that you can read all for free from the world's brightest minds in AI. So whatever you need to learn around generative AI, it's on our website at

your everyday AI.com. While you're there, sign up for our thanks a million giveaway. All right, before we get into this topic today about the correct way to browse the internet and the correct way to bring in up-to-date accurate information inside of chat gbt. Let's first go over the AI news. So Salesforce has partnered with Anthropic to enhance AI capabilities inside of Salesforce with its clawed model. So Salesforce has announced a partnership with Anthropic to integrate Claude models into Einstein 1 Studio, marking a significant advancement in AI powered business applications.

This collaboration is noteworthy. as it provides Salesforce customers with enhanced options for AI capabilities, aiming to improve operational efficiency and personalization across various sectors. So Salesforce customers can now use Claude3 models, including 3-5 Sonnet and Claude3 Opus and Cloud 3 Haiku within their applications. So specific applications for, you know, kind of this AI infusion here inside of Salesforce's Einstein. It includes, I mean, a lot of things, but generating personalized content, that's going to be the big one, summarizing customer service cases and analyzing contracts, which can

significantly streamline workflows and improve outcomes. So this integration, if you're looking for it, it's actually facilitated through the Bring Your Own large language model or B-Y-O-L-L-M feature inside of Einstein One Studio. So you have to make sure you have a subscription to that and have that enabled, and then it allows users to connect Claude models via existing Amazon Web Services bedrock environments. All right. Next piece of AI news. The Department of Justice has issued subpoenas to NVIDIA amid antitrust concerns. So the Department of Justice has taken significant steps to investigate NVIDIA's market practices, raising concerns about the company's dominance in the AI sector. So the DOJ has recently.

issued those subpoenas to Nvidia and other companies signaling a deeper investigation into potential antitrust violations. So this follows earlier questionnaires sent to those companies, marking a shift towards a formal complaint against Nvidia. So antitrust officials are worried that Nvidia may be making it difficult for customers to switch to competing suppliers and could be penalizing those who do not exclusively use its AI chips. The investigation also includes NVIDIA's planned acquisition of Run AI, a company that specializes in software for managing AI computing resources. So following these developments, NVIDIA's stock experience a 2% dip early last week and then a

significant nine and a half point dip on Tuesday, the overnight session, falling below its 50 day and 21 day moving averages. All right. Last but not least, new details have emerged about opening. AI's next model from two separate conferences, one in Japan and one in Taiwan. Yeah, both of these happening in the last couple of hours. So Open AI Japan CEO to Dow Nagasaki showed details of OpenAI's next model categorized under the heading GPT next on his slide at the KDDI conference in Japan.

So his slide revealed that the new model is projected to increase performance by a factor of 100 compared to GPT4. Yeah, compared by a factor of one or sorry, improved by a factor of 100. That's wild. Also, some more future open AI model specs were leaked. At the CEO summit in Taiwan, a corporate president and head of memory at Samsung revealed on a slide that open AI's GPT5 model may have 3.5 trillion parameters or about double the parameters of the current. class GPT4 models. So this comes after, obviously, recent reporting from the information showing details about this new strawberry mode or model formerly called QSTAR inside of ChatGBT that's currently being reviewed by federal regulators and will reportedly be used to create

synthetic data for training OpenAI's next model. Presumably that would be GPD 5 right now codenamed Orion or Orion. So, all right. It's a lot, a lot of AI news for today. There's always more. Make sure to go to your everyday AI.com and sign up for the free daily newsletter. And hey, live stream audience, love seeing y'all today. So thanks for joining us, Brian and Chrissy and Tara, Rolando, Marie, Cecilia, Kobe, Jay. We got a lot in the house today. Milka, our friends from YouTube, Michael and Big Bogeyface. Thanks for joining in. So here's the thing, y'all. I'm always one that brings receipts, all right? And when we're talking about chat GPT and getting up to date information from the web,

it's an extremely important topic to cover. This might seem like a small detail, it might seem like a footnote, but on the contrary, y'all, this is actually one of the most important things you need to understand when using a large language model, right? So we've taught literally, I think our numbers up to like 7,000.

We've taught like 7,000 business leaders prompt engineering live inside of chat, GBT, with our free prime prompt polish PPP course. All right. And this is one of the three biggest mistakes that people routinely make. Also, hey, if you've taken the course, live stream audience, shout it out, or if you want access, just type in PPP. I'll send you the registration link. We hide it.

You won't find it, you know, on our website unless you have the link. So this is one of the three most common mistakes that people make is assuming that the information that they have inside of chat GPT is correct, especially when they say, oh, well, it's using browse with Bing. So speaking of receipts, more than a year ago, I said do not use browse with Bing. It is terrible. And to be honest, at the time, it was. But over the last four months, Open AI has quietly rolled out some very impressive updates to browse with Bing. We've covered those literally step by step piece by piece in our newsletter and on our YouTube channel, which, yeah, I know the YouTube channel is a little cringe sometimes. But we cover up-to-date stuff on there, so you should be subscribed and go check it out.

But essentially, about four months ago, they didn't announce this. Open AI didn't announce this. And I put out a video and I said, hey, you know, we've heard all these reports. And this was four months ago. We've heard all these reports that Open AI is releasing this new search engine product, you know, aimed at Google. You know, now we know it's called Search GPT. OpenAI has started to roll out beta access to it. It might be a while before there's a general release.

But even before that, I said, y'all, no one has noticed this. But browse with Bing has been wildly updated. Under the hood, OpenAI didn't announce anything because essentially it is an answers engine right now. Yeah. So if you like perplexity, if you use chat GPT and browse with Bing correctly, you can get that very up-to-date accurate information. And y'all. So we go through a process in our PPP course called Refine Q. And a big part of that is the F in the I, which is F. in the I, which is Fectual.

Catching insights because you are working with a very old knowledge cutoff when you are working when you're working with chat GBT. So that's something again you don't like we don't talk about it a lot, but it's a problem. And this is why you now need to use Browse with Bing. Well, you should have been using it especially for the last four months, but we're going to detail it. All right. And the biggest reason is we need to get around the October 2023 knowledge cutoff date. All right. Here's what that means. Unless you're properly using Brows with Bing, and we're going to go over the pros and the cons, all right, here live on today's show. Unless you're properly using it, you're rolling the dice, right? You're rolling the dice to danger. You're rolling the dice to I might get a very bad output. All right. Because, you're rolling the dice.

all right because if you are not tapping into browse with Bing you are working with data y'all it is september now right unless you're listening to a replay of this in the future but it is September right now October 2023 is almost a year old now and that is the absolute best case scenario so to oversimplify this the way that large language models including chat gbt work is they essentially scrape the open internet yeah there's a lot of uh you know copy right aspects that are going to be getting litigated over the next coming months, quarters and years, right? So we'll save that for another day.

But essentially, these large language models are trained on the entirety of the open internet, closed internet, copyrighted pieces of works, et cetera. But the absolute freshest data inside of chat GPT is October 2023. But that doesn't mean that everything in there is October 2023. So now think, think of all those things. let's say most of it, best case, is just from the year 2023, but probably from earlier in the year. So you're probably looking at a lot of data that you don't know.

And it's generative, y'all. So every time you're going to get something different, you're rolling the dice with a large language model, unless you know, unless you know the tips and tricks, we're going to give you the weighted dice, you know. We're going to do that today on today's show. But otherwise, you could be working with data that's 18 months old, 12 months old, right?

At the very best case scenario, you're working with data that's 11 months old. And I'm going to go out on a very sturdy limb and say that if you're using chat GPT for business, I'm guessing nine out of 10 queries that you're putting in, you don't know where, or sorry, when the information is from. And I'm going to go ahead and say on another very sturdy limb that nine out of 10 times it's going to be extremely out of date. And you don't know. And you're just taking that.

and you might have a false sense of security, and you say, oh, this is good. This is, this is fresh information. I can use this. Wrong. It's in almost all cases out of date. And when you're working with out of date or old information, it makes it inaccurate. And one thing, large language models, well, you could call it a feature or a bug, but literally in their system prompt, they are made to be a helpful assistant, right?

Helpful assistant. So if Chad GPT or other large language models aren't 100% sure about something, and mainly that uncertainty is caused by, hey, if you're not giving it enough information, what happens, you know, we talk about this in our free PPP course a little bit, is these models become overly verbios, right? And they just spit out a bunch of buzzwords at you. And you read it and you're like, wait, did this actually answer my question? It's not because large language models are dumb. that's because you don't understand the knowledge cutoff. And if you're asking for up-to-date, accurate, recent information, a large language model can't provide that.

So if a large language model does not feel super confident that it's able to answer a question, it's just going to give you these vague statements, these blanket statements. So let's address that. So number one, or sorry, that's number one that you need to get around that knowledge cutoff. And by doing so, number two, you are ensuring your outputs are accurate and hallucinous. nation-free. That's the biggest thing, y'all, because, yes, large language models can hallucinate. They can make stuff up. And what, and the reason behind that is because you're not starting your chats, right? We, we preach and teach this concept of creating expert chats. And a big part of that,

like, like you would, training a new employee. You need to give them the most up-to-date information first, right? Before you're asking a model for an output. You need to, go through a training process and give it the most up-to-date accurate, accurate information first. But the good thing now is you can do that with Browse with Bing. Whereas six months ago, a year ago, I literally said on this very show, do not touch Browse with Bing with a six-foot pole. It's bad. It's very good now.

You should be using it. Let's dive in. So, and hey, live stream audience, let me know. I'm sorry. Am I still frozen, y'all? Hopefully not. Got to love tech problems on a live stream. Speaking live stream audience, let me know, are you using browse with Bing?

Are you calling out to it or are you just seeing that chat GPT is using it? Because there's a big difference. All right. Let's dive in. So here is a live example. All right. For a podcast audience, I'm showing some things on the screen here. I'm going to do my best to read them and explain them so you can understand what's going on.

So I am in chat GPT. I'm in the paid account, right? And I am in the latest version. This, even if you have a free account, this is still going to be relevant to you. All right. So all I'm saying is tell me about LLMs or large language models, right? And then obviously within a fraction of a second, chat GPT spits out. some good content, right?

So here it is. It's talking about foundation and architecture, neural networks, parameters. Great. All right. Fantastic. So we're getting some good information. One thing, though, you should always be doing, all right? And I need you to train your eyes for this when using chat GPT.

And shout out to Open AI because this is how it should be done. You always need to be looking between your query and the response from OpenAI. And you'll see right here, there's nothing in between there. Okay. So what that means is when I'm saying, tell me about LLMs and chat GPT spit something back. That means it's pulling from its knowledge, right? It's pulling from its training data. Here's why that's important. All right. We're going to take this concept. Again, this is very simple. Normally, you know, if I was using this for real purposes, I would go through prime prompt polish, right? If I was creating a research assistant. I'm doing this for simple explanations today, y'all. So next prompt. And each time I'm showing you a new example, I'm doing this in a new window as to not muddy up the context or impact the output that we might be getting from chat GPT. So now I'm saying use browse with Bing and tell me about large language models. So now I'm specifically telling chat GPT to use browse with Bing.

And then you'll see right underneath. So I said, remember, you need to train your eyes because a lot of times chat GPT is instantly going to spit out a bunch of great content super fast. And what do we all do? We all immediately start scrolling, right? and you almost start scrolling as fast as chat gpt start spinning it out. Hey, admit, live stream audience. Is that what you do?

That's what I do as well. But between your query and chat gbt, you're going to see this little icon, right? And it popped up here because I said using browse with Bing. So I'm calling out to chat GPD. So before I dive into this a little deeper, I want to talk about GBT4. Oh, it is GPT4 Omini. So Omni model. All right.

So if you rewind to more than a year ago, and this is one I used to say, do not use Browse with Bing. Browse with Bing was a dedicated mode inside of chat GPT. They eventually, you know, made it into the default mode. But it used to be its own dedicated mode. So if you wanted to use Browse with Bing before you started a conversation, you had to select. it from a drop down. And then all of your conversations, essentially, they were querying the internet. Okay. And I should explain Open AI has, and this is amazing. This is actually an amazing feature now that it works how it should work as an answers engine. So they have this

because of their partnership with Microsoft. So Microsoft has reportedly invested $13 billion for 49% stake in Open AI's profit arm. All right. So they have. They have. have access to use Bing. It is very quick, sometimes too quick. So previously, it was its own mode. So now you need to do what's called a call. Sometimes chat GPT, depending on your query, will decide on its own that it's going to use browse with Bing.

And like I said, this is why you need to train your eye. Because even when you are not making a call to browse with Bing, there's many instances in which chat GPT will automatically browse with Bing, which might be a good or a bad thing depending on what your query is and depending how that browsing works, right? And I'm an SEO geek too, right? I've been in the, you know, in and out of the SEO game actually for, I don't know, 15, 10 to 15 years pretty, pretty easily. Not counting back when I was a teenager and, you know, I think I made my first website when I was, how old was I? 12 or 13, right? And I learned very early on how to game

the SEO systems, but this is back in like GeoCity days, you know, back in the late 90s. But Browse with Bing, it's both good and bad because it's essentially it queries something. And depending on that query, it might bring you back bad information. And like I said, you may or may not even know or realize when ChatGBTGBT is using Browse with Bing unless you train your eye. So don't just scroll down to the bottom to quickly see. oh, let me see all this great information that opened that chat, chat, GBT just did for me.

No, you first need to train your eye because even if you're not calling out to Bing, it might use Bing without you even noticing. So you need to look for that little, it's going to say searched blank sites. All right. Yeah, more on that here in a second. So in this instance where I said using browse with Bing, tell me about large language models. I see that it's searched five sites.

So that means we have up-to-date and accurate information, right? It used browse with Bing. We're safe. We don't have to worry whether it's 12 months old, 18 months old. Correct? Nope. Because when we hit that drop-down, when we hit that drop-down button right there, so when it says searched five sites, that's actually a toggle.

And it is toggled up by default, or you do not see. any of the information. So you go have you have to go click that toggle that says searched five sites. And look what happened. This browse with Bing thing. And again, y'all, generative AI is generative. It does weird things sometimes. You can run this exact same prompt using browse with Bing, tell me about LLMs. And even the queries that it sends to browse with Bing, they might they might be different each and every time. So unfortunately in this run, here's what, it's sent to Bing. So it sent a query to Bing. So all I said, using Browse with Bing, tell me about LLMs. It's sent to Bing, LLM's recent developments,

2023 large language models. What happened, y'all? I just got a very false sense of security. Because number one, most people just completely scroll past and they don't even see or realize that they use Browse with Bing. Number two, even fewer. people click that button and look at the sources. Think of how often you or your team are probably using chat GBT, right? Yeah, that's right. If you're a manager, a director, and if, if, if, if you have a chat GBT ban or you don't have an official, you know, policy in place, guess what? All your, not all your employees, many of your employees are already using Chad GBT, co-pilot, et cetera. So yeah, your employees.

are doing this, so you should probably understand how to do it the correct way. So that's bad, right? That is a bad thing that we just got there, because I said using Browds with Bing, tell me about LLMs, and it's queried into being LLM's recent developments, 2023 large language models. Here's the thing. ChatGBTGPT is aware of today's date. That is one thing it is aware of. And if you run this same query multiple times, and you do you do that you.

different windows. You're not going to get this every time. Sometimes it might put the correct date. All right. And here's the thing with SEO, y'all. These, some of these, so some of the results we're getting are okay, right? So the first one is Hugging Face 2023 year of open LLMs. So it's like, okay, that's an okay result. And then we have something from Open Science 2020. So it looks like we're getting a lot of these like yearly recaps for 2023. But then we also have, uh, something from infoq.com. And here's the thing with SEO. So many times, someone's going to be ranking an article from 2020, 2021.

And here's what they do. They do some little tricks. They change something in the title. They change something in the meta description. But they actually don't update the content. That is a dirty little secret. So so many of these things that browse with Bing and other perplexity, et cetera, might not be bringing in the information that you want,

especially when you have more of an open-ended kind of browse with Bing call. All right. So now let's keep going on. And Marie has a question here. And y'all, if you have a question, please get it in. So she's saying, Jordan, did your prompt need to be more specific? Oh, absolutely. Marie, it's like, you know my slides.

Mind reader. All right. So we get bad information. All right, because here's just a zoomed in version, right? But we're getting these things. Some of them are coming back. Like I said, I ran the exact same thing again. And I got different results.

So in this time, it even searched September, 2023. I did it again and it did give me 2024. So this is what we need to figure out with targeted browse with Bing calls. All right. If you've done any of our free trainings, if you've heard me on this podcast before, you've heard me say targeted Browse with Bing calls. Here's what you need to know. And this is the solution.

So just like Marie said, we need to be more specific. So here's what I'm saying next. I'm saying using Browse with Bing, tell me recent developments about large language models from August 2024 to September 2024. So yeah, I'm not going to put in just September 2024 because it's only been a couple of days so far in the month. So now this is a targeted browse with Bing call. So what that means, like we talked about, you need to call out to browse with Bing. Because if you don't, if you just say, tell me about large language model developments,

it may or may not use Browse with Bing. And if it doesn't, it's bringing in old information. And then if you do make a generic Browse with Bing call, you're probably going to get old information. So we need to do a targeted browse with Bing Call. And now you'll see, and for our podcast audience, we're getting better results because now I see that it's binging. Did that ever catch on, y'all? Binging? I don't know.

Bing? You bing it? Google's a good verb. Adobe just introduced an entirely new way to create, bringing the power and precision of its creative suite into one conversational experience. Meet 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.

Are we binging? Yeah, maybe. All right. So now it's binging. LLM developments August 2024 to September 2024 news. So this time it decided to add the word news, even though I didn't tell it to. We're previously it didn't, but that's okay. So now I can see the different sites, right? So, all right. And it looks like most of these are from 2024. Not bad. So we got improved outputs. All right, but I noticed I don't really know these sites, right? Pastel.network, right?

There were some other ones. It's like, what if you don't know them? What if they're not authoritative sites? And if I'm being honest, going through these four different sites, that's not a lot, right? Like I said, what if two of the four are just old SEO sites, right? And they've essentially been the same since 2020 and they're just doing some, topical updates and the information is actually pretty old and pretty bad, right? You need to do a little more research.

Here's where we turn chat GPTs browse with Bing into an answers engine. Here's where we turn it into a legit perplexity competitor. And unfortunately right now, search GPT, it's almost like a separate product. It doesn't connect kind of asynchronously to the rest of the chat GPT ecosystem, which is unfortunate unless you're listening to this right now. That's why you tune in to everyday AI. We give you cheat codes. All right.

We're taking into step further. Now we're doing a highly targeted Browse with Bing call. So here's what we're doing. Using Browse with Bing. Tell me recent developments about LLMs from August, 24 to September 2024 only. Please search eight or more different sources, making sure that while using Brows with Bing, those eight sources are from eight different domains. Please go step by step, researching deeply, taking your time, and report back only once you've

researched the topic thoroughly and have the best possible view of recent developments about LOMs from August, 2024 to September 2020. So we're throwing in a little bit of, you know, chain of thought prompting here, right? We're telling it to go step by step. We're telling you it exactly what to do, almost in like a chronological. order here, right? And now you'll see, and it's not always going to do this, right? I've found y'all kind of the magic number is eight. If you ask for 10 or 11 or 12, sometimes it's just going to give you five and give up, right? It feels like literally, if you ask for too many,

chat, GVT and browse with Bing, it's like they're, you know, a defeated kid. They're like, oh, I can't do it. I did three. But if you ask for about eight or nine, usually it's going to give you, you know, seven or eight. That's a pretty good job. But I've noticed through our testing, that's probably the max. And that's okay, because that's a pretty good number. So now you'll see when I asked that, it went through and it found nine sites. All right. Tara said we be binging. All right.

I guess we're binging, y'all. So now we found nine sites. And we see the query that we binged, that we beat bing in. All right. So it said LLM's recent developments, August, 2024, or sorry, August, September, 24 AI. So even the terms that Browse with Bing is sending through and querying are changing a little bit, but that's okay. I would have liked to see August 2024, September 2024. That's okay, though.

We can do that for another day. Maybe we'll have an advanced course on using Browse with Bing. All right, but here's what we see. Now we have nine different sites, and they're all from different domains like we ask. because sometimes if you don't, it just gives you the same results from the same domain, you know, very similar information. So if you want a really good overview of, as an example, large language model developments, you want to be able to see a wide variety of sources.

But something else, let's say you're using this for more academic purposes. And one thing I notice when doing it this way, even with a very targeted browser with Binkal, we're giving it a date, we're giving it some kind of chain of thought prompting. All right. If you don't know what that is, don't worry. It's just kind of think of how humans think, right? If you have a complex problem or a multi-step problem, chain of thought just essentially means breaking it down into smaller steps, right?

So when we're assigning Brow-Bing to go do this, we're kind of giving it steps, right? And we're saying, hey, do this. Look for this query, go slowly, take your time, go different domains, right? So we're breaking the problem down and we're giving specific instructions. unfortunately, it looks like it brought up a bunch of blog posts, right? These listicles, you know, list of 20, list of the 21 best large language models. And it's like, okay, that's not necessarily what we wanted if we're asking for recent developments in large language models. We don't necessarily want, you know, blog posts and listicles and, you know, these, these, uh, entities that just

have great SEO, right? It's probably not what I wanted. So, If the sources aren't very good, you take it a step further. All right, I'm not going to read this whole thing, but here's what we added to the end of this query. So we added now, please only use authoritative sources for information, such as, and then we're giving examples of some authoritative sources of large language model paper, scientific research, etc. So I'm saying such as ARXIV, Microsoft, OpenAI, Google DeepMind AI, Kagle, and others. And then again, saying, please take your time and search at least eight sites. Now you'll see it's search seven, but that's okay. Because now I'm looking and it's only searching these authoritative sites.

It's looking at Microsoft. It's looking at DeepMind. It's looking at Brookings. It's looking at AR, AR5 ID. It's looking, right? So it's only giving me now more authoritative sources. All right. So that's great.

Guess what, y'all? We just solved so many of your problems. Whether you know it or not. That is how you should be using chat GPT. Okay. Again, you should be taking our, our free prime, Prime, Prom Polish PPP course and we go over,

we go over this a little bit. in our priming phase when we talk about Refine Q. Because one of the biggest things that you need to do, and, you know, obviously a trend of 24 has been RAG, retrieval augmented generation. That's essentially when you connect your data into a large language model, you do some fine-tuning, et cetera. So you can't technically do that in the front-end version of, you know, chat gpt.com. We kind of call this process, though, mini-rag, right?

And refine Q, that's our process of you take a 1.8 trillion parameter, large language model, and you need to make it smaller, right? And you do that in the beginning by giving it more up to date, accurate, and relevant information. Because now I can have a conversation and I can direct the outputs better, knowing that number one, I'm probably not working with data that's 18 to 12 months old. That's problem number one. Number two, I have much more control. I have a granular level right now of control over where of the rest of the conversation goes.

Now that in the context window of this chat, chat GPT understands exactly what I'm looking for. It has more up-to-date information. So it's not going to give me this, you know, wishy, washy, overly verbios content that says nothing, right? if you put in the work up front, you're going to get better results. All right, good, good question here. And if you have a question, get it in now. I'm going to be wrapping this one up soon. I don't need this to turn into an hour long podcast episode.

Where to go, where to go? Okay, Marie asking Jordan, I noticed in most of your prompts, you ask the LLM to take your time. Why is that? Great question. Great question, Marie. So there's been plenty of studies. that show when you tell chat GPT to take your time, take a deep breath, use manners, right? You get better outputs.

There's technically been research that show both sides, right? There's been research from world-class researchers. There's been examples. I believe that when you talk chat GPT in a friendly manner, when you tell it to take your time, take a deep breath as an example, it performed about 40% better on solving some complex math problem. So, you know, kind of encouraging it to think in chain of thought, to take its time, et cetera. Then there's been other research that says it doesn't matter. So I'm going to go on the side of, I don't know, my mom raised me to be nice, right? So whether I'm talking to a human or a large

language model, right? Antigdotally, I say it works much better. That's just me. All right. So here's the last thing about the correct way to give chat, TBT, access to the internet. Right. Because if you are putting, and I see this a lot, this is a big mistake, if you're putting a URL into chat GPT and you think, oh, well, it's browse with Bing, it's going to find the correct information. No, it's not. All right. So if you say as an example, right, and you see the little browse with Bing thing flicker and you're like, oh, it went to this, this webpage.

No, it didn't. So all it does when you give chat GPT, if you say, hey, you know, go research these specific sites or go bullet point. give me the takeaways from this specific article. You need to look at it, right? Click the drop down and see if it actually went there. But all it does is it bings or it queries the keywords in the URL, which may or may not bring up the actual web page you want. So it's technically a way that you might be bringing in hallucinations and not knowing it. So all it does is it bings the keywords in that URL, which may or may not actually be bringing

up the page that you want, at least when you do this inside of chat Gptee's default mode using Brows with Bing. Easy way around that. Use an internet connected GPT. So in this case, I'm using a GPT called WebPilot because a couple of our other GPDs we use a lot, we're being a little finicky this morning when I was planning this show, you know, using VoxScript, web reader, et cetera. But there's a lot of internet connected GPs.

So what this means is you need to, GPTs are free to. to use, even if you have a free chat GBT account, if you, if you want to build a GPT, a custom GPT, you have to have a paid account. But if you use an internet connected GPT like web pilot, then I can say, right, so I pulled two different URLs about some LLMs, right, because that was our whole topic, our example. And I'm saying, please bullet point the main points in these articles, right? I just dropped two, but you could drop, I'd say most GPTs do a pretty good job handling five, at least five, maybe six URLs. And then it's going to go through an internet connected

GPT. It's going to autonomously visit, right? You don't have to do it one time for each URL. Like I said, most GPTs can handle somewhere from four to eight. So if you have a bunch of research that you want to do inside of chat, GBT before getting your chat started, and maybe you don't want to leave it up to the discretion of Bing, even if you're using a targeted browse with Bing call, right? So maybe there's some long industry papers that you want to look at. Maybe it's something on your company's website, maybe a competitor's website, maybe a market research trend, etc. Right? And maybe you're like, ah, how am I going to trigger this with a targeted browse with Bing call? So an easy way to do that is to instead first use a GPT. So you need to go

into the GPT store, install it for the first time. And then after that, all you need to do is click the at button on your keyboard. That's called an at mention. You start typing the name of the GPT. and that's it. And then when you're done, grabbing that information and putting it into the context window, you can click the X button and then go back to kind of talking

with the default version of chat GPD. All right. Y'all, that's it. Now you know the correct way. And like Ben just said, what a Ben, good to see you. Ben just said Wednesday gold,

barabum, barabing, thanks. No, thank you. Thank you guys. But also, you need to put this into practice. You need to go out, do this for yourself now. So very quick recap. Huge problem if you are not careful when using large language models,

even chat GBT, that now has a great Browse with Bing feature. You always need to be looking with your eyes to see because chat GBT, even when you tell it, even if you don't tell it to, it might be going to browse with Bing. Bing. So you always need to scroll up and look, did it go to browser Bing? Yes or no. If it did, did you want it to? And also check out the sources that it's that it's going to. And then do a targeted browse with Bing call. Make sure you steer the model in the direction of the information that you want. And then you need to put that information into the context window. So then do this in the beginning when you start, when you're having a conversation, when you're having an

assignment, right? So it's literally like if you were to be working with a consultant. You always want to know, hey, consultant, do you know the most up-to-date information on whatever I'm using chat GPT for right now? Because if not, you're going to get old, out-of-date information. You're going to get this overly verbios content or output that says nothing. And worse, you're probably going to get some hallucinations if you aren't doing this the right way. But if you listened, if you went through today's show, you got it all right. All right. I hope that was helpful.

Hey, Marie said terrific session today at AI University. Fred said great session. Y'all, if this was helpful, do me a favor, right? Sometimes I stay up super late or wake up super early, right? I think just I've covered about every hour on the overnight clock except like 2 to 4 a.m. Right? I'm up at all hours. researching things so you can be the AI leader at your company. It takes us hours sometimes to do a

single episode, y'all. It takes you seconds to go ahead, click that repost button. All right. So if you're listening on, you know, Twitter or LinkedIn, go ahead, retweet this, repost this, tag a friend who needs to hear this. That would be great. If you're listening on the podcast, we appreciate your support. Please like, subscribe, follow. Can you like? No, subscribe to the podcast. Leave us a rating if this was helpful. Make sure y'all, we're going to end this probably next week. All right.

Go check out our thanks a million giveaway. Go to Your EverydayAI.com. You're going to see a little banner at the top of the website. You sign up for the giveaway. You're going to get a referral link. So if this was helpful, go send that to people because you can win either a year-long subscription to whatever your favorite large language model is, chat, GPT, you know, co-pilot pro, you know, consumer license, y'all, not paying for your whole company, one consumer license.

So, you know, chat, GPT, Gemini, Claude, et cetera. And we're also giving away some consulting sessions. So go ahead, get that unique referral link. Go send it to your friends. Go post it on social media. We're wrapping this up soon. We're going to give one prize away to the winner and then a couple random prizes. So even if you can't catch the winner, all you got to do is go enter.

And then all you got to do to keep growing is keep tuning in. Go to your EverydayAI.com. Thank you for tuning in. We'll see you back tomorrow and every day for more Everyday AI. Thanks y'all. 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. Stay in 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.