EP 25: Decoding Lottery Patterns with ChatGPT-4

Transcript for this Episode:

INTRODUCTION VOICEOVER: This is Tech UNMUTED. The podcast of modern collaboration – where we tell the stories of how collaboration tools enable businesses to be more efficient and connected. With your hosts, George Schoenstein and Santi Cuellar. Welcome to Tech UNMUTED.

GEORGE: Welcome to the latest episode of Tech UNMUTED. We've got a bit of an interesting podcast today. We're going to take a look at some functionalities in ChatGPT. Really, ChatGPT-4 that lets you do some data analytics. It's going to be a little bit of fun we're going to have here. We're looking at some lottery numbers from the state of New York. We're going to walk you through and show you what ChatGPT does with it, how to get to it, that kind of stuff.

I do want to start by reading a disclaimer. That disclaimer is before we dive in, the data discussed in this podcast is based on historical lottery results and is for entertainment purposes only. Past lottery trends do not predict future outcomes. Every lottery draw is a game of chance with each number having an equal likelihood of being called regardless of the past data. Do not base your lottery decision solely on this information. Always play responsibly. Understanding the odds are significantly low that you will win. We accept no liability for choices made based on our content. If gambling becomes a concern, seek help promptly.

SANTI: There you go. In other words, do not use this podcast-

GEORGE: To pick your numbers.

SANTI: -to pick your numbers. [laughs]

GEORGE: I'm going to walk through-- Let me get my screen up here and share this screen.

SANTI: You know what, this is going to be fun, actually. This is great.

GEORGE: This is GPT-4. This is the paid version. Specifically, what I'm going to look at here is advanced data analytics. The reason we're doing this, this is a fun thing to look at with the lottery, but this is some really cool functionality to do data analytics. A lot of stuff that's in here is what you'll see with Microsoft Copilot when it gets into GA. This is a paid service. I think it's $20 a month, is what I'm paying for this. Again, this used to be called Code Interpreter up until a couple of weeks ago. Which was an odd name, and now it's got a better name.

The data I'm going to look at is Mega Millions' data that runs up until about 2020. I think it goes back 10 years. I'm going to load the data set in. The place I found this was at data.gov. There's a bunch of stuff we looked at, but we chose this one, which is beginning, actually, it goes back to 2002, and then it goes up to 2020. I'm pretty sure it's where the data ends. I will point out to you, you have a 1 in 302.6 million chance of winning the Mega Millions. This is fairly difficult.

Once you're in here, you pick GPT-4, you pick advanced data analytics, you go to your file sorter. Here, I've got the lottery numbers. It's going to load that spreadsheet in. Once you've got the spreadsheet loaded, you can query anything. Very simple query is just to say, "What do you see in this data that is interesting or unexpected?" We'll do that as our first query. We're doing this live. If you look on the left-hand side, you can see we already played around with this once before, but let's see what happens. It's going to start--

SANTI: This is great because you're using what's called-- This is natural language processing before our eyes. We're using natural language to give this bot instructions. Not only is it reading your instructions, but it's comprehending. This is what AI is all about. We're looking at it right before our eyes. This is great.

GEORGE: It takes a similar process no matter what data you load. It starts to look at columns and rows and tries to figure out what's going on. It identified there's a draw date, there's winning numbers, there's a multiplier. Interestingly, it started to identify, hey, numbers 1 through 26, it seemed to be drawn more frequently, less frequently 27 through 39.

SANTI: Very interesting.

GEORGE: There is an interesting element to this that we flagged, and I'm going to prompt it on that in a second, but we'll see if it figures it out itself this time. It has identified the numbers go 1 through 69. It's analyzed the data. That's the one thing I'm going to flag back to it. It's going to give us a chart.

SANTI: Oh, look at this. It's giving you suggested reasons for why you're getting certain results. See that?

GEORGE: There is a reason. I think that gives you frequency.

SANTI: Interesting.

GEORGE: It is fairly odd in that chart that there is a drop-off. I'm going to prompt it. It's still going. It's still analyzing, it's analyzing.

SANTI: This is amazing because we're seeing that a lot of people hear of ChatGPT, but they never really had a chance to play with it or use it. We're actually giving the audience here a live view of this artificial intelligence analyzing the data and answering your question. By the way, it's still answering your question. You said what's interesting and what's unexpected. It's giving you the answer. It's crazy. Oh, look, it's talking about the multiplier now.

GEORGE: That's the additional-- You can pay extra to get the multiplier rate. Clearly double. The payout is more frequent than 10 times. Yes, you would expect that. I'm not a big lottery player, but I guess that would be an expected outcome. There was a thing I'll note in the chart that was a little further up, and I'll scroll back up once it's done analyzing. There was a perceptual-- you could see an actual drop-off at about 60 for the numbers that are drawn.

SANTI: It said something about that it looks like numbers were added later on. It said that in the description. That's interesting.

GEORGE: I'm going to go back and say, reanalyze real quick. Taking out what happened at that point in time. Also, I'll ask it to identify when the number 60 or higher had started to appear in the numbers. Now it's doing-- This is far more analytics than we saw the first time. We asked a very similar question. It is giving-

SANTI: Oh, look at this.

GEORGE: -number 7 and 15 have been drawn together 20 times. These are some interesting things that absolutely did not identify this before. It's got other sequences of numbers drawn together. Again, this is all random from the standpoint of a lottery draw.

SANTI: This is amazing. Look, for example, it would take us, if we took the same data and try to do the research ourselves, it would take us probably days to try to reach all these conclusions that the AI just did. This is the beauty of the latest advancements in artificial intelligence. Look at all the analytical points we have now in a matter of minutes. This is great.

[silence]

GEORGE: Now there's something that happened with 60. Again, I didn't go back and actually look this up, but it was clear that the number 60 and above didn't exist at one point. I'm asking it to go back and look again, tell us what's different. I'm going to say, in the end of this, because this took a really long time, I'm just going to tell it, make the summary of your analysis shorter this time. I'm going to send it back to do this.

SANTI: I'll tell you why it's doing this. I did see that it did capture-- it said something that it appears like those numbers were added later. It was able to find that, hey, something about the data. Apparently, these numbers were added later. It was able to pick that out. Oh, there you go, September 26th of 2020.

GEORGE: There's clearly a pivot that happened here. I suspect there was another one prior to that because there was an inflection point in another place, maybe around the number 50, something like that. 60 was only drawn after 2020. Again, it did not figure this out the first time around when we analyzed the data, but it has the second time. These are completely independent uses of this analytics.

SANTI: Yes, I think, a lot of it has to do with the prompt too. Slight changes or slight differences between how you prompt it will give you a slightly different result sometimes. That's just with AI in general.

GEORGE: This got down to a much narrower dataset because I think this data only goes through the end of 2020, so it got really narrow now. It's all 60 come in, 60 and above. Now it's slicing down at a smaller level below that. Now it's going to do a little analysis on what it sees in this data, and then we're going to ask it to pick some numbers for us. We'll see what it comes up with.

SANTI: Look at the first, number one, it says numbers 27 to 39, which were previously drawn as Powerball numbers, seems to be drawn after they added the 60 plus. Interesting.

GEORGE: Interesting.

SANTI: [laughs] For some reason, having the additional 10 numbers or whatever that, at the end, that gap, that space in the numbering sequence fell off. That is very interesting.

GEORGE: How do we want to ask this? It's not going to want to give us the answer, but I'm going to say, "What are the most likely numbers to-"

SANTI: Be drawn.

GEORGE: "-be drawn next?"

SANTI: Oh, very interesting.

GEORGE: "Give me three ways you would come up with this."

SANTI: You're going to really make this AI think.

GEORGE: Yes. It's not going to want to answer. I'm sure it's going to warn us, going to do. It's apparently uncertain since it's designed to be random. However, there are some strategies.

SANTI: Look, it even called it hot numbers. [laughs]

GEORGE: Yes.

SANTI: You see that? [laughs] And cold numbers. Oh, okay, there's actually a terminology for this, and it's actually giving you that. All right. Wow.

GEORGE: Again, the value of this overall is you could take survey data, do the analytics. You could step through all of the data. You could then even say, "Summarize the results that I would use in a report. Summarize the results that I would use in a press release." In fact, I might try that at the end of this to say, "Write me a two-paragraph press release to talk about frequently used numbers or numbers to pick for the lottery," but let's see what it comes up with. If we were to go with hot numbers, gave us a couple hots.

SANTI: Did you notice that it gave you the hot numbers, but they're not in numeric order? They're scattered. You see that?

GEORGE: They didn't give six. Let's go back and ask one more query here. I'm going to say, "Give me the last three bullets, but make it six numbers in each." See if it understands that, because that's a little bit of a complex query actually, if you think about it.

SANTI: If you're listening to this podcast on the audio platforms, you're probably going, "What is going on?" [laughs]

GEORGE: It's got to be painful.

SANTI: You're going to have to come to YouTube because this is just mind-blowing. This is one of those YouTube-only podcasts. Listen, I understand all the advancements in AI, I know how we've gone from machine learning to natural language processing to conversational AI to even neural networks. I understand all those concepts. This is great. To see it actually unfold before your eyes this way, it just it never ceases to amaze me. I'm having aha moments every time I see it, and I don't get bored of seeing it. That's just what's so fascinating. Look at this.

GEORGE: It is interesting if we look at this. It understood what I said. Most frequently drawn numbers, hot numbers, gave us six.

SANTI: Yes.

GEORGE: Cold numbers, gave us six. Common pairs, maybe not as useful. I don't know. Maybe it is useful. You might like a pair, it's up to you, but it gave that data. Let me do one last query here. I'm going to say, "Write a two-paragraph press release about the last response you made that highlights how ChatGPT can help you pick lottery numbers." Let's see what it thinks about this. If you don't use ChatGPT, one of the things you need to be aware of is don't just set it loose if you know you need something long or concise. It will tend to do long and oftentimes very long multi paragraphs of stuff.

SANTI: It's part of your prompt. You gave it some parameters. You said, "Listen, keep it within two paragraphs." Otherwise, it will go along. That's what's great about this, is the prompt is key. All right. There is San Francisco, California.

GEORGE: Yes, this is pretty funny, actually. If you read this, "OpenAI is advanced language model. ChatGPT is offering a new perspective on how to approach the age-old challenge of picking lottery numbers. With this vast analysis capability, ChatGPT has demonstrated the ability to sift through historical lottery data to identify frequently drawn numbers, less common ones, and even pairs of numbers that often appear together." I'm not going to read the rest of it, but it's a little funny.

It's almost like it gets it's funny. It does do the note again on the end. You'll get this on a lot of things with ChatGPT. It gives you a warning. If you try to do something that it's not-- doesn't have a data set to do or something like this that you can't really predict, this is completely random, it's going to caution you on this. Again, we thought this was an interesting thing to look at today. We've used it a couple of times with some survey data and other stuff.

This data set actually is one of the better ones I've used to do analytics. It was really clean and straightforward. It only had a couple of columns that have multiple rows of data. It was both more simple on the front end and lengthy in the data set. I have found with some survey data, if you get traditional spreadsheets, cross tabulations, whatever it is back on it, they are more challenging than this to analyze. I've had to restructure the data to get it to work. It has identified some insightful things that I would not have otherwise identified with the data.

SANTI: Or it would have taken you an extremely long time to reach those conclusions. One thing I would like to do is I'd like to go back and look at your prompts because I think the key to all this is what your prompts are. Your very first prompt was very short but to the point because you challenged the AI. You said, "What do you see in this data that is interesting or unexpected?" Now it's really, it's going to take those two words to heart. It's going to say, "Hold on. I need to sift through this data and come up with interesting points and the unexpected."

Man, did it come up with data. This is a significant amount of analytics that it gave you. If you scroll down, just we're not going to read it all, but if you scroll down, there is a lot of massive points here.

GEORGE: Interesting things.

SANTI: Yes. A chart.

GEORGE: It gave a chart here in return that helps you understand what was going on. The one caution I will give you, we don't load any proprietary or sensitive data into this. I'm not 100% sure what the privacy policy is on the back end of this. We've not really drilled into it. We've just taken the stance from our standpoint that we're not putting sensitive data in here. We just don't know where it goes or how it gets ingested. Something we probably should look at in the future on a podcast.

SANTI: At some point we have to review that. Not only that, I'm assuming that once you go into a paid model or something, that's part of what you're paying for, some type of policy sets. Let's look at your second prompt because once it generates all this--

GEORGE: It did all of this, and then I said, "Look at it again." This was the first one. I got a little more detailed. I knew there was an issue at the number 60 because we'd seen it when we did a dry run on this previously. It came back and said, "Let me identify." I think it came back last time and said something like, "I'm sorry, I missed that." Then it re-analyzed. This time, by the way, to Santi's point, somewhere up there, it apparently did see that there was some anomaly.

SANTI: It did catch that. It looks like numbers after 60 were added later. It did identify that.

GEORGE: I don't really have a full understanding of how it's regenerating and its approach on the back end, but it is loosely possible that that previous go-round we did, it learned, "Hey, I missed something, and I need to not miss it the next time." This goes back to the broader premise with AI. This is a tool to help you do your job better. It's not going to replace your job, and it's proven here.

It missed a key point, or in the second time, it didn't totally miss it, but the first time it totally missed the key point, which was something changed at a point in time with the numbers.

SANTI: Yes, this is just fascinating, man. No matter how many times-- Oh, here's the other prompt. Correct. “What are the most likely numbers to be drawn next?” I love that prompt, and “give me three ways you would come up with this.” Then it came up with its own description, hot numbers, cold numbers, and then just an analysis. This is great. What I love about it, it's actually describing its approach as well.

It's telling you, hey, here's how I'm going to develop your answers. I'm going to look at these things, and then it does it. It explains what its approach is going to be. Then as you scroll down, it gives you the output, and there it is.

GEORGE: Then it didn't give me all the numbers. I said, "Hey, give me the six."

SANTI: Give me the six.

GEORGE: This is a little bit of a complex prompt, even though it's a sentence. I'm saying the last three bullets, they weren't actually-- they're technically bullets, but they were numbered.

SANTI: It understood. Remember, it's comprehension. That's the thing about AI. It comprehends what it's reading, which is scary. Then, finally, the press release.

GEORGE: Press release, which is funny.

SANTI: That was hilarious. Wow.

GEORGE: I can say, "Give me a chart of this data to be used with the press release."

SANTI: Oh, nice.

GEORGE: See what we got. That's a pretty good job with the charts. Cut and paste them out. You don't have the data set. You can ask for tables as well. You could get a table and cut and paste it out and drop it into Excel. Taking a little more time here. It must be thinking about what it's going to do. Let's see what happens there. It's looking at the data sets.

SANTI: Oh, look at that. Oh, yes, I see what it's doing.

GEORGE: I'm wondering if it understands the press release words.

SANTI: Well, what it is, it's going to create a graph based on the three things: hot numbers, cold numbers and common pairs. There you go. There it is.

GEORGE: There's the chart. It's interesting. It's not great. It's interesting.

SANTI: It's okay. It gave you a visual of the hot numbers, the cold numbers, and then the common pairs. Listen, for anybody who plays the lottery and likes to have a scientific approach to gambling, I guess, this is good data. Listen, people have been doing this since the lottery existed. There's people who analyze historical data to try and come up with their next pick. AI just made that process shorter for them now. Oh, what are the odds [crosstalk].

GEORGE: What are the odds of winning based on the numbers in the original data set? See if it can figure-- This is fairly complex because it has the data set, it's probably-- We intuitively know there's a set of numbers that I think go up to either 69 or 70 in this dataset. You could do some-- A way to prompt it as well would've been-

SANTI: Yes, scroll down to generate it.

GEORGE: -put some random draw of 6 numbers out of 69. What's the odds of picking that at any given thing? To determine the winner-- It thinks it's Powerball. By the way, it's Mega Millions. Number range 1 to 69. It's pretty interesting.

SANTI: Oh, it's giving you the math formula. Hold on. I get it.

GEORGE: Let's see what the odds are. The other thing we saw on the web said like 302 million. Let's see if it comes up with the same number. It's saying it's even more challenging. 1 in 438 million.

SANTI: Wow.

GEORGE: Again, it thinks it's Powerball. I don't know if Powerball is different than Mega Millions.

SANTI: It's fine.

GEORGE: Lottery is not my thing. Hey, so with that, we'd love to get some feedback on this. If there's other data you'd like us to take a look at and analyze, do another podcast to do the same thing.

SANTI: By the way, entertainment purposes only. [laughs]

GEORGE: Clearly.

SANTI: That's the key. That's the key here. We would love to showcase stuff like this. Because this is really the only way to understand the advancements of artificial intelligence is to actually see these types of things unfold before your eyes. I say let's do it. If somebody has an idea for a data set that we could use on ChatGPT, leave it in the comment section below, and we'll entertain that. I think we could do another episode using suggested data sets.

I would love to do that. George, you hit the ball out of the park for our 25th episode. I love this. This is great. Folks, this does bring our podcast to an end. Please go ahead and subscribe. Subscribe to our podcast on your favorite podcast platform and as well as on YouTube. This one was definitely a YouTube episode.

GEORGE: That's definitely YouTube.

SANTI: Folks, until next time, remember this, stay connected.

CLOSING VOICEOVER: Visit www.fusionconnect.com/techunmuted for show notes and more episodes. Thanks for listening.

Episode Credits:

Produced by: Fusion Connect