368 - How to Use ChatGPT Effectively ... And Not Get Fooled!

0:00How to use ChatGPT effectively without getting fooled by it. So you've heard me say before that technology is a powerful servant but a terrible master. Well, maybe you haven't heard me say that before, but I've said it before. So, well, that's especially true with a tool like ChatGPT. I do use ChatGPT a lot for analyzing data, occasionally writing something, building some templates, even analyzing some patterns across mentees. So it's an incredible accelerator for thinking and productivity, but, and this is the key, it's not infallible. So like any tool, it needs a smart operator, which in this case is presumably you or maybe a VA virtual assistant.

0:52So now let's talk about how to use ChatGPT effectively and what to watch out for. First, understand what it is and what it isn't. ChatGPT doesn't know things in the way humans do. It predicts text based on patterns it's learned from data. Some of those are accurate, and some of those clearly are not accurate. That means if you ask it to summarize a report or a list of names, sometimes it creates what sounds right but isn't. And those are what we call hallucinations when we talk about ChatGPT. Confident mistakes that look completely plausible. So, for instance, if I upload a series of reports, and each report has a name on it, and I ask it to do whatever I ask it to do, sometimes it comes up with names which were never on those reports.

1:51Totally fabricated, totally not malicious. It's just AI trying hard to fill gaps that it thinks should exist. How that happens, I have no idea. Why it comes up with names that are not mentees, I have no idea as well. But it does, and so understand it. So what's the takeaway? You can certainly use ChatGPT as a drafting assistant, but not an oracle. It can get you a pretty good percentage of the way analyzing P &Ls, analyzing EBITDA numbers, analyzing balance sheets, maybe 80 % of the way. But you, the human, hopefully, must provide the last 20%. Verification, judgment, and context. There are some things you see.

2:43If you ask it to create a letter, you will see it almost uses dash marks instead of comments. Why? I don't know. It almost always starts off a letter. I hope this letter finds you well. I hope this email finds you well. And certainly when I look at those and receive those, which I get all the time, I know, okay, this is something that ChatGPT put together, usually with very little input from the person. So don't be those guys. So here's how I've learned to minimize mistakes and get better results. One, really important, be specific in your prompts. The clearer you are, the more accurate the output will be.

3:25Instead of saying, for instance, summarize a report, maybe you want to list the names that appear in that report. And then don't infer or add names not listed. Use that prompt, if you will. So ask ChatGPT to verify itself. Ask it to double check that all names on the data are from the uploaded file. I ended up having six source PDFs of reports that were on industry reports. And I asked it to create graphs, and it did that. And I looked at the graphs, and they made no sense to me. And I said to it, where did this data come from? And ChatGPG said, I don't know. Now, I had provided the six reports, so if I had said to it, only use the data in these six reports, no other data, maybe I would have gotten a different answer.

4:28So the important point here is prompts and what you ask it to do are very, very important. So if you ask it to do things like that, that small line forces a cross-check and reduces hallucinations. And use it iteratively. Don't settle for the first answer. Refine it. Say, that's good, but show me the sources, or reformat that. I had it analyze a letter from a mentee, and it gave me a bunch of good comments. But I said to it, the first paragraph I like, the second paragraph seems that it has no transition and just jumps from paragraph one to paragraph two. At which point, chat sheet said, excellent feedback.

5:17I'll change that. So remember, you need to interact with it. But know its limitations. It's not intended to be your private data. data. It's not connected, better way to say it. It's not connected to your private data unless you upload it. It can't see in the emails that you've written, at least I think it can't, unless you integrate those tools directly to it. So don't expect it to remember something unless you provide it again. Oftentimes, you'll think, well, it's got a history here. And sometimes if you're in the conversation, it will remember that, but sometimes it won't. So don't be bashful to ask it to analyze something and give it the data that you want to be sure it analyzes.

6:08Okay, so how about versions? So you might say that ChatGPT or ChatGPT5, which version is better, and each version is smarter, faster, and better reasoning. But even ChatGPT5, which I'm using now, although my marketing director prefers ChatGPT 4, can still make mistakes. The main difference is that ChatGPT 5 has a broader context window, meaning it can read and remember much larger documents at once. And it's much better at following your tone and logic, and I certainly see that. But regardless of the version, the same rules apply. Trust but verify. AI can certainly enhance your thinking, but it should never replace it.

6:55So here's a challenge for you this week. If you're using ChatGPT or thinking about using it, treat it like an apprentice. Give it clear instructions, check its work, and learn how to get the best from it. Because when used properly, ChatGPT can save you hours, sharpen your communication, and even help you think more clearly. I recently asked it to generate 110 steps, mini steps, in a format that I could use with mentees from the Acquisition Advantage Bootcamp. And it did that in, it took a lot of going back and forth, but probably an hour. And someone who looked at it said, you did this, Bruce?

7:42I said, yeah, with the help from artificial intelligence. Well, that's miraculous. Given your lack of technology training and coding, to have it produce this, which is really good, is pretty amazing. And one of the mentees said, you could have done that in two to three hours, Bruce, if you had two to three years of coding. And I could see it coding off to the side. So it was pretty impressive. So for me, it did something that I could not have done myself. That was clear. So absolutely. But remember, so go ahead and take that challenge. Because when used properly, it can certainly increase your productivity.

8:31But when used blindly, it can lead you straight into confident, well-written nonsense. And it does have hallucinations. So keep your judgment sharp and use technology like ChatGPT or Claude or Perplexity or any of those to amplify your intelligence and your talents but not replace it. So how about an action step? Use ChatGPT or the AI of your choice for one real business task this week. Maybe you want to have it draft an email or summarize a document or review some financials. Any of those things you can have it do. And then compare its output to your own version. Notice where it helps and where it makes mistakes.

9:22That awareness is how you will go from being a tool user to a tool master. And I'll give you an example of that. One of my mentees had an 80th birthday and asked me to be the emcee. at his three-day gala and to give a speech. And my marketing director said, do me a favor, Bruce, and don't let ChatGPT write your speech. You write your speech. And that was good advice. I did that. I didn't put it through ChatGPT at all. I asked my marketing director to review it, and he had some comments for it. But that was an important speech. It was one that I wanted to be in my tone of voice and based on my experiences.

10:08And it was well received. And that, for my marketing director, was good advice. You write it yourself. So don't take, like many will, the easy way out. Use it as a tool. Go ahead and say, how can it help me? But remember, trust but verify. You use it as a tool, not as a crutch, which many will. And you can certainly think of it, if you're anti-technology for whatever reason, is that it's just a progression of technology. When I first started this, we referred to what we did with shoe leather. We literally didn't have the internet. I know that's crazy for 95 % of the people that will listen to this.

10:53But I got my first email in 1995. So before that, there was no email address. And so you used the Rolodex, you used contacts, you didn't have any LinkedIn searches. It's much easier with those in business acquisition today than it was back then because the tools are so much better, and ChatGPT is just one of those tools. So I hope this is helpful. I hope that you use the tools, but remember, ultimately, you're the pilot steering this ship. don't let the autopilot take over and do everything for you.