Become A GenAI Power User In 30 Minutes with Mark Hinkle at AIE 2024

Welcome everyone to my crash course in Gen AI Desktop Productivity. In the next 30 minutes, I'm gonna teach you how to be a generative AI power user, uh, using chatbots like chat, PT um, CL philanthropic, Claude, Google Gemini. And I'm really excited about, uh, this 'cause this is the kind of stuff I do every day, and it's been hugely impactful for me.

So my name is Mark Hinkle. I run an AI consultancy called Perty Labs. Spent the last 25 years as an executive in emerging tech, and for the last two years I've really gone deep into generative ai.

Um, today I help businesses understand and use ai, um, to improve their productivity. And I'm gonna give you the benefits of what I teach lots of people to do, and I'm gonna do it for free. So let's get started.

All right, um, today, and this is from, uh, Microsoft and LinkedIn, they did a survey, and three out of four people are already using AI at work. Um, 75% of knowledge workers use ai. Um, and most of us, or almost most of us have, have, um, have done this in the last six months.

So there's not a huge corpus of people out there that have a lot of gen AI skills. So we're all learning this together. Um, 90% of those people that do use AI sales, it saves them time because they can focus on work, be more con creative, and the majority of them at 83% enjoy their work more.

So I know I do, and I hope you do as well. So, the easiest path to generative AI is to use chatbots. And the one that we're most familiar with is, um, open AI's Chat, GBT, but I'm gonna talk about some other ones today.

Um, and we're gonna, we're just gonna go through some of the, uh, strengths and limitations of each. And I'm gonna give you a power user tip for each one. And I think that if you apply these, these tips and understand that you can get, uh, lots of benefits from these, um, you may, may start using multiple models, uh, multiple chat bots to help you, or you'll at least understand the chat bots that are advocated by your organization.

Um, so, uh, let's get started. And we're gonna start with OpenAI. And, uh, we're recording this in May, 2024.

And just this week, uh, OpenAI made a lot of announcements and, um, it had to do with the model or the brain behind Chat g pt. And so before their best model was called G PT four, and it was very good at, um, analyzing text. And then they had another model called Dolly, and that's very good at generating images.

Now they've combined everything into a model called, um, GPT-4 O, and that's a, uh, lowercase o and that stands for, um, Omni as in omni Modal. So if you modes are, um, language, natural language processing, video and audio, and it really has been a huge, uh, improvement for a lot of people. Um, what it allows you to do now is have really good conversations with chat GBT, and that's in the text like you do on your browser or, uh, via the Open AI app on your phone.

And, um, it's really added a lot of personality and empathy or perceived empathy in these interactions. So what I use OpenAI for is it's sort of my workforce and it does a lot of things like, um, it improves my organization of thoughts. I can do research there, I can create outlines for, uh, writing pieces.

I can do tons and tons of stuff. Um, and it's gotten better in the last few days When I, when I put this presentation together, um, a few weeks ago, it didn't have near these capabilities, and I'm getting a little bit surprised with, um, every interaction I have there. Um, one of the things that you probably haven't done or the majority of people haven't done, but, um, uh, you can do, and this is my, my latest life hack is I use, uh, chat GPT with Siri.

And, uh, you can download chat GPT to an iPhone or an Android phone. But, um, apple has added in the last year these shortcuts and these shortcuts will allow you to, um, generate tasks from, um, Siri and other applications that are more complex. So you may go into your Siri and say, Hey, Siri, ask GPT, and that will launch OpenAI.

Um, and you can, you know, ask it questions and it'll re send responses instead of from Siri, from, uh, chat GPT. So you can search YouTube and find lots of ways to do this. You can create your own shortcut or there are pre-configured shortcuts that folks have put up on the internet that you can download.

But, um, if you, uh, in a recent interview with Sam Altman, the, uh, uh, CEO of OpenAI, he talks about the fact that he keeps his phone on his desk, and as he's working in his desktop environment, he can ask chat GPT questions via Siri, and it gives him another channel for information. So while you may be using chat GPT to generate, uh, text or images or analyze data, the sort of workhorse, um, everyday applications, it's just one more app, uh, way for you to, to leverage chat EBT. And there's some great folks out there that have amazing tips.

Um, one of our presenters today is, uh, uh, Nancy Bain. She's, she's a, uh, you know, a force of nature when it comes to how she uses, uh, chat EBT to run her business. Um, let's go to one that you may not be familiar with, and this is Anthropic, Claude and Anthropic.

Claude is a, uh, and you know what's funny? If you hear that little beep I said I had, I asked Siri to do something, I didn't turn off my phone. And, uh, Siri's talking to me, um, is, uh, so back to Claude.

Claude is a chat bot, just like chat GBT, but it has a different brain. It has a large language model, um, that is called Opus three at this time. And then, you know, by the time a few months roll around, they may have another name, but, uh, think about these models behind the chatbots as a librarian who's read everything in the, in the library and instantly can recall all that information and help you.

So Claude is, uh, also available for free and, um, you can use Claude to do all the things, not all the things, but a lot of the natural language tasks that you do in Chat GBT. But where I find it to be very helpful is as a copywriter, and you know, that Chachi bt often provides you these sort of unnatural sounding snippets of text that sound something like, in a world where generative AI is taking over and they use words like tapestry and delves and landscape a lot more than, you know, I do, at least, and a lot of people, I feel like Claude has a more natural language, um, tone, and that's where I use it. So sometimes what I do is I would actually take the text that was generated chat GBT, and copy it into Claude and ask it to improve my writing.

And that has been a huge advantage. It also gives me some more creative ideas. So if you think, if I go back to my librarian example, you know, if you have a librarian at Claude and you have a librarian at chat pt, they may have read different books.

So they might have a slightly different perspective. And that's what I really like is, is chaining those things together. And often what I'll do is take what comes outta Claude and I'll put it into my word processor, and then I'll use yet another, um, chat bot.

And this is Grammarly, which we has been around, uh, for a long time as a spell checker and grammar checker, but now it has AI capabilities and I can even use that to improve my writing. So, um, when I first started using ai, I thought I'd be able to tell it to write me a blog post or write me a report, and it would come out just as I wanted. And the reality is, it does probably 80% of the work really, really well, and 20% of it is where I really need to exercise some creativity and some oversight.

Um, so that's, that's Clyde. Um, there's a part of Clyde that is a really good resource that a lot of people don't know about it. And so when we talk to a a chat bot, we use prompts, and a lot of us have probably seen on the internet, I use this prompt to do this, or I use this prompt to do that, but it's not a skill that that is broadly around.

And so when we want to make prompts that do repetitive things, we wanna make sure that they're good so that we get a, um, good result most often. And these, these prompts will not generate the exact same output every time. These models are not what we call deterministic, which is, well, like when you put a an equation into the calculator, you always get the same answer.

But when you put a prompt into these large language models, it has some degree of creativity or what we perceive to be creativity, and it's actually a little bit of randomness that allows this to mimic the human creativity. So when you put a prompt in, you'll never get the same answer, but you would want to get the same general result. You want something that is written, um, you know, with the same kind of, um, style and consistent with the see.

And this is where, uh, anthropic, the company that delivers Claude has done a really good job, is they have offered a product, uh, prompt library to give you ideas of how to create these prompts and use these prompts. And then they give you a prompt console for prompt testing. So the Anthropic Console allows you to put your prompt in there and look at the results and then tweak it and see how it changes the output so that you can get a prompt that you can reuse.

For example, if you write a report every week on your status, um, and you can upload the transcript of the meeting notes, you may have it generate the, the, the report in the same format every time. That's where you might go and find the prompt, use that prompt in the workbench to refine it, and then save that. And I'll talk a little bit later about how I save my prompts, but if you reuse those prompts, a cut and paste into the chat, GPT or anthropic, and you can use anthropic prompts in chat GBT.

Um, but the workbench is really being tested against the chat GPT or against the Claude model. Anyhow, that's, that's sort of my pro tip. Once you get really good prompts for repetitive tasks, you can see huge productivity gains.

The next one that we have is Gemini. And Gemini came to us from Google it. If you've heard of Google Bard, that was the old name, it's now being been renamed as Google Gemini.

And, um, as we record this in the middle of May, 2024, um, Google IO just happened and they released a whole lot of updates. It's very exciting. Um, Google Gemini can be available for free, or you can pay for it as all of these, um, uh, chatbots are, and you get some additional features.

And rather than going into these features as they change every time, I'll, I'll give you the general features that I like for Gemini. And, you know, Gemini is done by Google. It has access to real-time information from Google, and that's the big advantage.

It does the, pretty much the same thing as Claude. It's not quite as robust as, um, uh, chat GPT, but it's very good. It's a good copywriter.

Um, it also has this idea of a context window, and I might touch on that a little bit later, but if you think about the context window is of the memory for a conversation is Gemini's big advantage is it has a larger context window. So if you uploaded a large report, it can keep its train of thought, whereas with the smaller context window, and of the three we've talked about so far, chat, GPT has the smallest of the three context windows. Um, you may have to break up your task for a context window so that it doesn't, um, forget what you told it.

It's just like their memory banks or your, your memory. If you have a long conversation with someone, you most likely will remember the last thing that they said better than what they said at the beginning of the conversation. So Gemini is a little better at, at that memory.

Um, and it's integrated with the Google ecosystem. So if you are using, um, Google Docs, Gemini can integrate with your docs in various ways. You can chat with your Google Drive and your spreadsheets, and you have your own Google copilot.

Um, in there, the same thing holds true for Microsoft. And I did not, um, I'm not gonna talk too much about Microsoft copilot today because Microsoft's copilot is, uh, powered by um, GPT for our open ai. And so it's interface is a little bit different, but the capabilities are pretty similar.

It's also, um, very much like Gemini integrated into Google. Uh, Microsoft's copilot is integrated into the Microsoft 365 suite. And so you might start seeing copilot showing up in your PowerPoints, in your Excel, in your word.

And that's, um, how that works. So, um, this is the thing that I think is the most interesting. So I have a lot of documents that I say to my Google Drive, and if I want to query what's in there, or I wanna write a report based on reports I've updated or documents I've saved, I like to use Google Gemini.

And the way I do that is, um, and, and things are changing fast. com, and I add my Google drive to the interface, and then I can chat with my data. And that's one of the things I think you'll see as a theme across a lot of these chatbots is that chatting with your data.

So having a large report spreadsheet, um, you know, bunch of emails, et cetera, and being able to have a chat to draw, you know, recall information from there and collate it and create new works from that. So that's, that's the exciting thing that, that we see from that. Uh, the next one is meta ai.

So meta, uh, formerly known as Facebook has a very powerful large language model called lama, and they have made that accessible through pretty much all of their, um, meta properties. That includes Facebook, Instagram, WhatsApp, and it allows you to, um, access AI right in your interactions in your, your sort of social mediums. ai, which is a, um, free portal that's much like, um, chat GBT, but very much geared towards their ecosystem.

Um, and so my power tip there is, and, and this goes for, I, I used WhatsApp, uh, as the example, but, um, you know, it's very good if you need to, if you're already a WhatsApp user or a Facebook Messenger user, you can chat, um, directly with the AI and get answers. So, um, you know, I have a, a Ford Bronco and it's a newer Ford Bronco, and sometimes the radio, um, which is all electronic and fancy gets stuck and I need to reset it. So if I'm on the road and I want that information, I might text, uh, via what WhatsApp to Meta AI and say, listen, um, can you gimme a description of how to reset the head unit in my Ford 2022 Bronco?

And it re applies with a procedure and it works really well. So, um, it's sort of that on the go AI is how I use that. Um, next we have probably one of my favorite, um, and unique use cases is perplexity.

And Perplexity is sort of, if Chat GT and Google Search had aaba, it would be called Perplexity ai. And what it does is allows me to, um, search for results and it gives me attribution to those results. So if I ask it what the 10 best restaurants are in New York City, it would create that list, but it'll also show me where the list came from.

So it might have come from Google Reviews or Yelp or Facebook or somewhere else, and I can understand where that data is coming from and decide whether or not that result is, you know, valid or not. So, um, that's how it works. Um, it does realtime web searching.

It's not as creative as Claude or Chat GPT or Gemini even. It's more of a research assistant. And that's how I use it.

So one thing I might do is if somebody, um, sends me a document and I wanna make sure that I, the document is correct, I may upload a document. In this example, there was A-K-P-M-G generative AI study, which I, I had a good sense was probably right, but I can upload that document, ask perplexity to fact check the document, and then when it finds facts in there, it puts footnotes in and those footnotes are linked to the sources. And you can, if you're doing that in a web browser, which, um, that's the most likely way that you would be doing it, um, you can actually click through and verify that those, uh, sources are good sources and not spammy.

And you can chat with the document, as I mentioned earlier, to actually create, um, to understand how those, um, the thought process behind that document and that those answers actually. Um, so that's, that's how I use perplexity. So that's, that's gonna end the portion on chatbots, but I wanted you to be able to listen to the first 15 to 20 minutes of that, take away five actionable things and, uh, start implementing them today.

Now we're gonna talk a little bit more about a skill that I think is, um, important for all of us to gain. It's, uh, it's a skill that that is, has a fancy name, prompt engineering, but what it really means is communicating with AI through prompts. And, you know, if you've used them, and I'm, I'm assuming if you're here today, you have, you've, you've generated a prompt, but, um, you've asked the AI to do something, create a document, answer a question, uh, create a picture, et cetera.

But just like when you have employees or children's, or even your pets, giving clear instructions is going to give you a better result. So prompt engineering is just giving clear instructions so that you get a better output. And so, um, I'm gonna go through some tips that I hope will, um, give you that skill in a very amount, short amount of time.

So, um, as I mentioned earlier, these, these models have a little bit of variability. So what you want to do is you want to reduce the variability that is undesirable and increase the variability that is desirable. And that's just a matter of creating a good prompt.

And here's an example of a good prompt is you want to structure it in a way that the prompt can follow instructions. So, um, you know, in this prompt I'm gonna ask it to write a blog post, and I'm going to give it a role. The role is, it's a copywriter and SEO expert, and then I'm gonna give it some constraints.

I want it to write a blog post, but I only want it to be an 800 word blog post. And then I give it some context. And the context is, I wanted to use SEO best practices, and I give it additional context that this blog is for marketing professionals.

And, um, I do the next thing, and this is one of the things that a lot of people skip in their prompt, is to ask it to provide some reasoning. If you can't, if you remind it that it needs to come up with examples based on real world examples, it's gonna ground that model and have it mimic some kind of logical progression in what it delivers. And then in this one, I gave it a criteria.

So I wanted it to be engaging, I wanted people to read my blog post. Um, you could provide other criteria that said, I want it to be, uh, very sober in tone. I want it to be, um, appropriate for children under the age of 18.

All those are criteria. If you start thinking in this, this sort of framework, you're gonna have a lot better results. Um, I touched on context windows earlier, but the way those context windows are measured are in tokens.

And so chat GPT doesn't really understand the language you put it in. What it does is it takes that language and it tokenizes it or creates discrete bits from your words and then applies probabilities to that. Um, the chat GPT, um, has a limit of 40, um, 4,097 tokens, I believe, and the context window.

So that means that your text of a certain length will be remembered. Um, it's roughly 500 words now. That's what it was before.

So, um, this week we've seen some updates if it's, and it probably will get to be larger context windows, but just remember, you can't just stomp a ton of text in there and have it remember because it turns that you're 500 words into tokens. And then the tokens are limited by a context window, but that's not, you know, the end of the world. What you can do is when you have that much, you can chunk it, turn it into chunks.

So if you had an outline for your presentation or report and it was much longer than that, you could take it and chunk it out into 500 word pieces one at a time so that you don't over run the context window and then ask it to analyze your report. So, um, that's my quick and dirty way to do that. There are other plugins for your browser that are called, you know, chunking plugins that does that for you automatically.

And what it does is it breaks it up appropriate to that context window. So if you're a Chrome user, you can go to the Chrome store and find that and um, do that. But, um, as I said earlier, um, there are different size tokens, uh, context windows measured by tokens for different chatbots.

The other thing that you can do to make your prompts better is you can do, um, give it examples. So when, and, and there's a name for these different kinds of learning that the models do. And when you don't give it an example, um, you just give it a prompt.

Um, like you type in create a, you know, story about a dog leaving the pound and you don't give an example of the story. That's an example of a zero shot learning. But if you want to give it examples, um, you can give a one or two examples, and that's called Few Shot Learning, where you say, give me a story about a dog leaving the pound.

And you give it an example story in your prompt that will show the, the model what the right kind of output is. Um, and then the, the one that's many shot is if you gave it 10 examples. com, um, social media post, and I wanted it to be about, um, you know, a tip around how to use chat GPT better.

I might provide five tips or 10 tips in my prompt so that it knows what they look like in the format that I would want it. So 280 characters for X, and then it will look at my examples to inform its outputs. So that's called Many Shot Learning.

So here's an example. Um, this is a little bit of an eye chart, but if you look at what I did here is I used, um, a type of formatting called Markdown, and it's just a way to, um, convey formatting in a piece of text. So Markdown sometimes is written and then converted into a rich text format or into Word or HTML.

So I took that original type of format where I had a role and objection, instructions, format, et cetera. And then I added examples and I read it right in markdown. So, um, by writing a markdown, it helps to organize the prompt and make it more understandable.

Um, I also added these examples below here so that, that it would know what I wanted it to look like. So I gave it the role content creator. I, um, said the objective is to create a tweet and I can even be more specific that promotes an event, shares news, increase awareness, and then I gave it instructions, and I did that in a ordered list.

So, uh, specify the topic, I said to use clear, concise language, um, include hashtags, ensure the tweet is engaging, and then I gave it a format that said it shouldn't exceed 280 characters. And if it was applicable, a call to action. And, uh, then if you see here, I said, now let's start by terming the specific topic you want to tweet about.

Um, if I don't answer that in this format, I, it'll actually prompt me, especially in chat GBT, but that's the sort of variability that I keep this as a, um, template. And then I use my examples below. So that's sort of what a high quality prompt would look like in the way I rate my prompts.

Um, and then when I have a good one, I just save it in a library. So I use Notion you could use Word, um, there's lots of ways to do that. So I, I, I chose, um, notion initially, but then, um, over time there's a number of, uh, extensions for Chrome that came out and I, I use them more often now.

So, um, two of 'em are ai, PRM, which is you pay for, and then Prompt Forge, which is free and you can pay for additional features. But what I do is once I get that prompt into a format that I want to reuse it, um, I can save it in there in the AI PRM, it shows up in the background of my webpage as an overlay with all my prompts. Um, and it has a library of prompts from other people that I can use.

The same thing with Prompt Forge, it's a little different. It floats on the right hand side of chat, GBT, and I can open a file cabinet and then I just, um, can either create and save prompts or I can run them and then inserts it right into the, um, chat GBT interface and, and these, uh, prompt managers as time goes on. Um, I believe that, uh, prompt Forge supports other, um, things than chat GBT like, uh, Claude.

So that's pretty much an overview in 30 minutes of how I think you could be really more, you know, twice as productive as you were previous to Gen AI without a lot of work in 30 minutes. Um, every week I share tips like that. I create a prompt of the week in the newsletter.

Um, you know, same name as the conference. You're already signed up if you attended here. But, um, the AI enterprise that I owe for productivity chip strategy and, um, you know, I hope you subscribe and stay subscribed.

Thank you for your time.