Unraveling ChatGPT's Power: An AI Experiment with NFL Score Predictions
Posted on October 6, 2023 by Fusion Connect
Watch & Listen
Tech UNMUTED is on YouTube
Catch up with new episodes or hear from our archive. Explore and subscribe!
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. Today we're going to take a look at some additional functionality in ChatGPT, specifically GPT-4 in the paid version. We're going to take a look at a couple of plugins. Similar to Episode 25, I'm going to read a disclaimer upfront. Before we dive in, please note, the data discussed in this podcast is based on historical football scores and is for entertainment purposes only. Do not base your football wagers on this information. We accept no liability for the choices you make based on our content. As always, play responsibly. If gambling becomes a concern, please seek help promptly.
SANTI: In other words, what you're saying is, we had so much fun with the first attempt at using ChatGPT to predict lottery numbers, that we decided why not just take it to the next level and do NFL scores? That's where we're at, right?
GEORGE: Let me share my screen here. I'm going to bring up the web browser again. Again, this is in the paid version.
GEORGE: You're going to go over to GPT-4. We looked at advanced data and analysis beta on the last one. Today we're going to take a look at the plugins. The two plugins that I chose on here are two browser apps. What they do is allow you to pull in current data. To take a look at this, I already pre-built the prompt here. Based on everything you know about previous NFL scores, what do you think the scores of the games for this will be for week 3, 2023 which is the data that's at that link I have there under schedules? It's just live and it's written.
SANTI: Oh, okay. Okay, okay. Hold on. What you're doing is, you're linking so that it can go look at the upcoming games, right?
SANTI: Then because of the plugin, it's able to pull historical data for NFL scores?
GEORGE: Well, no, the opposite. I think it knows historical scores, but this lets it look at an actual live current website.
SANTI: Got it.
GEORGE: The database is old. The database is a year and a half old or whatever max. [crosstalk]
SANTI: Ah, yes.
GEORGE: This allows it to pull in and compare against current data.
SANTI: Okay. This is cool. All right.
GEORGE: Yes. Again, this is useful for, even if you're building content, it gives you the ability to pull in some current context on trending stuff that's going on. I can't get it to do at least a really broad search, but I've done it with specific websites and it works pretty well. Let's see what happens here the first go through here.
Our plan on this by the way, so you might see us wearing different clues at the end of this video. Our plan is to run this, and then do a comparison on Monday or Tuesday, and see how accurate the predictions are. I did run this, you can see it on the left-hand side, I did run it previously. I'm an Eagles fan. The first go round when it predicted the scores, it predicted the Eagles were going to lose to Tampa Bay on Monday night.
SANTI: You weren't happy about that.
GEORGE: I'll be interested to see what it thinks this time.
SANTI: Just for the folks who are watching. Again, if you're listening to this podcast on one of the audible podcast platforms, this is one you want to go to YouTube for to watcha> as the AI goes to work. We are recording this technically on Friday. It's Friday, September 22nd. By the way, I am wearing my red shirt. For those who don't know, I have a son who's deployed in the military, and red stands for, "Remember everyone deployed."
It is Friday, and it's red shirt Friday. We're going to come back [chuckles] and give you an update, and we'll append it to this recording on Monday. This is going to be fun honestly. This is going to be a great episode. All right, so what do we got?
GEORGE: It's interesting that first go round, it gave a much more detailed view of this. It drilled down into every single game, and it ended with, but it didn't give scores, it just said, "I see these games." It said, "Do you want me to tell you the scores?" A little different this time, so I'm going to say, “Based on historical data, tell me what the likely scores are for the games this week that were at the link I provided.”
GEORGE: Let's take a look and see what this does.
SANTI: Natural language process. I love it. [silence]
GEORGE: Historically games are close, so odds are it's going to give us some fairly close scores. We know there's blowouts every week and that kind of thing.
SANTI: I love the narrative. It gives you an explanation. Here are some of the patterns; close games, high scoring affairs, defensive battles, home field advantages, even looks at that. This is really interesting. Injuries, it takes into account injuries. Look at this. [laughs] This is great.
GEORGE: Now it's forgotten what the specific matchups were. Let's go back and we're going to copy this and bring it down to the bottom and say, the week 3 matchups are here. Let's see if it figures it out. It's interesting, as I said on the other one, it took a completely different approach with the exact same prompt. It had identified that it was week 3 in the data.
SANTI: That means it's--
GEORGE: This is now using WebPilot and I'm-
SANTI: I see that.
GEORGE: -not sure which one it used in the previous one. Now it's going down the path it went down before. It is likely just going to identify the games-
SANTI: Right. I see that
GEORGE: -initially. Then let's see if it gives us a prompt to try to give us a score. Again, this has many uses. This is a fun example we're doing. Again, don't rely on any of the information that we get at the end of this with some scores predicted. You can use this for a number of things to get live data access using these plugins.
SANTI: The question I have is, will the Dolphins beat the Broncos?
GEORGE: Let's say, I think it hopefully ends the same way it did before, and says, "Hey, do you want me to actually predict the scores based on what I know?" Which it's just going to know historical data. It's going to know, I would assume home field versus not, and are they in the division?
SANTI: It's giving you some insight now.
GEORGE: Yes. Interesting, but not scores yet.
SANTI: Look at this. I know, but look at the analytics. It's giving you some historical insight. Oh, look at this. This is fascinating.
GEORGE: Then we'll work through the end of this. There are-- the thing to note on the plugins, there are a lot. It's more than a hundred, it's likely more than a thousand. There's all different plugins.
GEORGE: Yes. I chose these two. There's a popular tab or something. They were in the popular tab, so I just plugged them in and I've played around with them a couple of times.
SANTI: Look at all this. Look at the narrative.
GEORGE: Yes. Gave a little bit of this the first time around when I did it offline, and also gave the score. Let's see where we get the score. Philadelphia, let's take a look here. It still thinks Tom Brady's playing for Tampa Bay, so its dataset is old.
SANTI: [laughs] It's interesting though.
GEORGE: I gave it the current scores.
GEORGE: I'm going to say--
SANTI: It's interesting that it pulled up old data even though it's using a plugin that should be grabbing new data, so I found that interesting. Okay.
GEORGE: Now I'm going to be deliberate. “Tell me based on historical data you have, what the likely scores of these games will be?” Hopefully we get a score here. Speculative, understood.
GEORGE: Doesn't really want to do it. Here we go. Now we're getting scores.
SANTI: We are getting scores. It says [crosstalk] a range.
GEORGE: It picked up the Thursday night game from last night and predicted a score. We know that game already took place. It's O for 1 at the moment because I believe San Francisco won 30 to 24, 30 to 20, something like that. O for 1 so far. The remainder of the games it is giving us a score. We're going to let this thing run through, and we're going to stop for the moment and we'll come back next week and do an analysis on Tuesday or Wednesday and see.
SANTI: I want to see the Miami Dolphins and Denver. Denver versus Miami, potential score could be 23-20 in favor of the Broncos. I'm not a fan of ChatGPT right now but that's okay. [laughs]
GEORGE: Let's see- [crosstalk]
SANTI: We'll see.
GEORGE: -if we get to the Eagles yet because the Eagles game is a Monday night. It would've been further down. Let's see.
SANTI: Yes, there it is.
GEORGE: Especially with Tom Brady, the score would be around-- I'm going to do one other prompt, and then we're going to close this out just to be selfish. I'm going to say for the Philadelphia versus Tampa Bay game, Tom Brady is no longer playing for T. Bay. Knowing that, what would the score be?
SANTI: You misspelled what?
GEORGE: Let me go back there and-
SANTI: There it is.
GEORGE: -fix that.
SANTI: All right.
GEORGE: I'll do one final thing here. Just being a little selfish to make myself feel better.
SANTI: Look at this narrative. Oh, a potential score could be 27-24 with a slight edge to the Eagles.
GEORGE: I feel better about that one.
SANTI: Let's what happens with the Denver Dolphins game. [laughs]
GEORGE: With that, we're going to close out the first portion of this video.
SANTI: Yes. This is the first half, the first half of the recording.
GEORGE: We'll do a second half and do the prediction. We are not going to do this and release any of this stuff ahead of a week. As we said, we don't want people relying on any of this information and place wagers. This is clearly just for fun. It doesn't really have a lot of data around anything, it just knows previous scores.
SANTI: Listen, here's what's fascinating. Same exercise we did with the previous podcast. This is artificial intelligence making predictions. It's a predictive AI module. Of course it's based on historical data, but still, think about the narrative it gave us. Think about how it reached all these different conclusions, how quickly it did it. Man, AI is just amazing. We do this all the time you and I, but it never ceases to amaze me. It's just fascinating. This is going to be awesome.
GEORGE: Think where we were a year ago. A year ago, you had some nascent AI tools out there. People started using ChatGPT like--
GEORGE: Q4 really last year. It was not great. It started to get a little better. GPT-4 came out, it was not as good for the first couple weeks. It learned; it got better. This you wouldn't have been able to do a year ago in any real way as just an individual user going out. We're going to cut to the break. We'll come back. I suspect we'll be wearing different clothes. I don't know.
GEORGE: We'll do another recording of this. We'll take a look at what the actual results were, and see how good ChatGPT did. O for 1 to start, because it had a wrong prediction for the game that took place last night, which I suspect at that link it had the score by the way and it still predicted the wrong score, so it didn't figure that out.
SANTI: Yes, maybe.
GEORGE: Let's take a look at the other games and we'll see you guys on the other side.
SANTI: We'll be back in a few.
GEORGE: We're back.
SANTI: Yes, we are and man, nobody saw that coming. [laughs]
GEORGE: I know. Just the Eagles were amazing, weren't they?
SANTI: Yes. Can I mention the Dolphins won 70-20? Sorry Broncos. That was crazy. That was probably the craziest game ever. Anyway, let's pull up the results. People are here because they want to know if AI can possibly predict a football game. Listen.
GEORGE: Let me share, I--
SANTI: You got the numbers.
GEORGE: Yes. I dumped it all into a spreadsheet. Before I actually hit the share button, oh, did I hit the share button? I hit it anyway.
SANTI: You did. Yes, you hit the share button. You're sharing.
GEORGE: I'm going to continue with the disclaimer that was at the beginning of this video. This is for entertainment purposes only.
Clearly based on the results that are in this, no one would use this to use to support their sports wagering I guess, unless you went against everything that it said.
GEORGE: Let me break it down a little bit. We left the Giants in there even though the game had already taken place. It predicted they were going to win. They did not win. They were incorrect on that one. The only interesting one in here was the LA-Minnesota game was almost spot on on the final score.
SANTI: Off by one point.
GEORGE: A point away. Overall, it was 40, a little bit over 43, almost 44% correct in its predictions. We didn't do this against the spread or anything else. I don't really know if that would've made a difference. Again, keep in mind there's two data sets here. This was the broader data set of past game experience and everything that ChatGPT has access to. That's stuff that's a year, a year and a half old.
Then it had live visibility into the current piece of data I gave it which was the teams that were playing this week. I believe that also had the year-to-date record for each team. There's only two games into the season. You would've expected actually a little better outcome than this. I'm just guessing.
SANTI: I don't know. See, listen, you know what this all, this was a fun exercise. Same thing we did with the previous podcast, it's for entertainment purposes. Here's what it showed me. What it showed me is, no matter how intelligent AI is, and no matter how much data you can pump into an AI prompt and how much historical data, a game of chance is a game of chance.
I do not foresee any way that AI could possibly predict a game of chance other than give you some historical analytics, statistical outputs maybe. We just saw this past weekend that stats went out the window because the Dolphins won 70 to 20. Who on earth would've thought 70 points? I think it's probably the highest they've ever scored in a single game I think.
GEORGE: It's third highest actually because I looked it up.
SANTI: There you go.
GEORGE: Apparently, they had the ball with a minute or two left and could have scored again. If they had scored, they would've set the record for the most points ever. The previous, the most recent time that somebody had exceeded 70 or 69 points, whatever the number was, was 1966.
SANTI: There you go. That's crazy. That was the lesson for me. It's like artificial intelligence is great at giving you historical output, historical analytical output, it's good at giving you statistical data. None of those things determines a game of chance. That's the reality because anything can happen.
GEORGE: The rubber's going to meet the road in a couple of weeks when the Eagles and the Dolphins play.
GEORGE: I think the game is in Philadelphia. Hope that it is. We'll see who the winner is in that, but there's only three teams left that are three and O. Maybe we do another quick revisit before that one and do a little five or six-minute run-through with ChatGPT, and see if we can pump a little bit more data in and see if it'll predict the outcome of that game.
SANTI: I can tell you we're going to win but that's okay, we can go through the exercise if you wish. Anyway folks, listen, this was just for fun. I hope you guys enjoyed it. It was a different twist to our podcast.
GEORGE: Again, it's really about the tools. How do you use--
SANTI: It is about the tools. Yes.
GEORGE: How do you use the more advanced features, in this case what's in ChatGPT, the analytics piece that's in it. Again, when we're using GPT-4, we're using a couple of the live plugins to do live web data.
SANTI: That's right.
GEORGE: Use writer, however, is analyzing the data that's there against its existing database using some of the analytics functions. If you're not using this, you need to take a look at it. This is a lot of the functionality that will be in--
GEORGE: In Copilot, yes, when it goes into GA. You're also seeing many folks who I think it might just be on corporate licenses, possibly fives at the moment but there's a preview of Copilot that's hanging out to the right-hand side now on mine for the last couple of days. It's more interactive to create some content pieces more than anything else. It interestingly, I was cutting an image of something to put it in a PowerPoint deck and I was on a call at the same time that I was doing that, and it said, "Do you want to drop this into the chat for the call?", which is a helpful step.
SANTI: Sure. It's one less thing you have to do?
GEORGE: Yeah, for sure.
SANTI: All right.
GEORGE: With that, I'm going to let Santi close this one out.
SANTI: This was fun. Thanks, George for putting this together. This one was your idea, so was the previous one. Listen folks, if you haven't done so, this is the good opportunity to do so, and that is to subscribe. Subscribe to this podcast Tech UNMUTED on your favorite podcast platform. Until next time, stay connected and stay curious. See you.
CLOSING VOICEOVER: Visit www.fusionconnect.com/techunmuted for show notes and more episodes. Thanks for listening.
Produced by: Fusion Connect
Tech UNMUTED, the podcast of modern collaboration, where we tell the stories of how collaboration tools enable businesses to be more efficient and connected. Humans have collaborated since the beginning of time – we’re wired to work together to solve complex problems, brainstorm novel solutions and build a connected community. On Tech UNMUTED, we’ll cover the latest industry trends and dive into real-world examples of how technology is inspiring businesses and communities to be more efficient and connected. Tune in to learn how today's table-stakes technologies are fostering a collaborative culture, serving as the anchor for exceptional customer service.
Get show notes, transcripts, and other details at www.fusionconnect.com/techUNMUTED. Tech UNMUTED is a production of Fusion Connect, LLC.