Greetings and salutations, folks. 

Welcome back to another edition of FridAI. I’m your host, Mike King, CMO of AIPRM. So, thanks for joining me again. I want to talk about live crawling and the concept of retrieval augmented generation today.

So as we all know or should know, ChatGPT has a series of limitations with respect to its training data, primarily that the data only goes up to, I think it’s September of 2021. So it’s not going to know facts, it’s not going to know current events, it’s not going to know things like the Nuggets won the NBA finals. It’s not going to know that Agent Orange has been hiding documents in his bathroom and he has a chandelier in his bathroom. Ridiculous stuff, right?

ChatGPT Training Limitations

So how do we get around that?

Well, there’s a concept called retrieval augmented generation, and it’s something that search engines are using for generating their content based on search results.

So effectively, what they’re doing is when you ask a question or you form a query, it’s going to then go get a bunch of those results and then use that for pretraining the language model before it responds. So think of it like this: you ask a question and then it’s like, “Okay, well, what are the top X number of results?” And then it’s extracting a bunch of information from that and then using that for the pretraining, and then you’re creating a big context window with all of those results, and then it’s giving you your response.

Retrieval Augmented Generation

So Google has a version of this called retrieval augmented language model, pretrained or Realm, and that’s what powers BARD and search generative experience. And then, of course, ChatGPT is using their own version of that as well. That seems to be based on Bing’s retrieval system.

So there are two ways to get live crawling or your own version of retrieval augmented generation in ChatGPT. There’s ChatGPT Plus, but it only works with GPT-4. And the thing about it is that it makes the decision of what it’s going to go after, whether it’s going to perform a search query or it’s going to go to a URL directly. So you don’t necessarily have all the control that you might want because sometimes it may crawl a series of pages rather than the specific page that you’re looking for. In this example that I’m showing you, it actually crawled that URL directly.

Two Ways to Get Live Crawling

However, with AIPRM Elite or higher, you can have it limited to just the specific URL that you’re going after, and then you have a lot more control over where that content is placed because the way it works with AIPRM is that we are extracting all that content from that URL and then putting it directly into your prompt.

So I get a lot of value out of this myself, you know, as a marketer. In this example, you’re seeing I put in a URL for a prompt that I forked for generating tweet threads based on a URL. And so we’ve got a series of different prompts with live crawling that you can use, and you can also generate your own.

Live Crawling with AIPRM

So when you’re generating your own, it really comes down to two things that you need to account for. One, the prompt in a live crawl prompt becomes the URL, meaning the prompt placeholder. And then you also want to put in crawl text to indicate where you want to place the copy from that URL in your prompt. So then you have control over how it’s handled.

Writing Your Own Live Crawling Prompt for AIPRM

That’s all I got today. I definitely think you should be trying out live crawling specifically to get better output. Ultimately, it helps with hallucination, it helps with getting stronger output that’s more related to the facts that you want to be talking about, and ultimately just yields a better experience in what you’re using ChatGPT for. So if you haven’t installed AIPRM, I definitely recommend that you do that. Also, in the comments, wherever you see this, feel free to give me suggestions on things that you want me to cover because my whole goal here is to just make generative AI way more accessible, way more usable. And I’ll see you next week on the next FridAI.

Check out the previous episode of FridAI, How to Write Effective ChatGPT Prompts where Mike explains why prompt engineering is crucial and how choosing the right words can significantly impact the output you get when using generative AI tools.

If you have any problems with the prompts that you’re writing, you just want some input or some feedback from folks, you’ll find what you’re looking for in our AIPRM community.

Think you’ve got what it takes to create the best AI prompt for small businesses? Submit to our contest for a chance at $5000 and tons of other cool prizes. Submissions end Friday, July 7th, 2023.

Get ready to unleash the power of generative AI for your business with our all-new Teams feature. Get your team aligned on generative AI without any restrictions, regardless of your AIPRM plan.