It’s week five in the world of AI Maddeness, and the Jets are a franchise history-making 4-0 to start the season. As it just so happens, in that other universe 25 years ago the NY Jets also opened 4-0 only to go 9-7 on the season and miss the playoffs. The real 2000-01 season was a heart breaker, but AI Maddeness is where dreams come true.
Week 5 is the Jets bye week, and it couldn’t come soon enough. There are quite a few things to work through before week 6 starts, what’s more I’ve become a bit obsessed so this break in the action allows me to take a breath. A few things to do:
- Style the AI Maddeness, AI Sports Zone, and Gangus Green sites. I’ve been promising this for weeks, but is this the weekend I’ll truly lock-in?
- Re-install the Madden 2001 ISO-Rom. The emulated game has been freezing up leading to all kinds of issues, so I need to try a re-install before week 6.
- Create a separate official Jets site for team news, injury reports, comprehensive stats, etc. This will be a model for having a site for every team (aspirational I know), but if I come up with a solid template it would be easy enough with WPMS.
- Another dream is to see if there’s any way to dig into the guts of Madden 2001 and get some of the stats out in an automated fashion that can be fed directly into OpenAI GPTs. Tom Woodward’s comment makes it that much more fun to explore such a far-fetched ambition.
- Play a few games against some other undefeated teams in week 5. In particular, the Jaguars against Steelers and the Bills against the Colts. Both the Steelers and the Colts are undefeated and are upcoming match-ups for the Jets, so it will be good to get an in-game sense of those teams.
- Get around to doing a highlight video of a game.
- Finally, testing out what it would be like to make an AI-generated podcast reflecting on previous week’s game and getting ready for the next week’s, the bye week may provide the perfect occasion to reflect on the Jets vs Steelers going into week 6.
If nothing else, the to-do list highlights that AI Maddeness can quickly become a pretty significant time sink. I do love it, but I find myself wondering WTF I’m doing after spending hours scraping screenshots for stats 🙂
One thing that might help in this department is the power of GPTs, or pre-trained chats that have all the instructions I’ve been typing and re-typing into chats stored so that when I paste in the screenshots with stats, it knows exactly what to do. Tom Woodward opened me up to this brave new world and it has already paid dividends. I have GPTs for scraping stats, creating specific graphics, writing game previews, etc. Instructions can be fine-tuned over time and with this setup I can actually imagine a world where stats, text, and images might be automated between ChatGPT and the blog. The holy grail would be getting the data directly out of Madden 2001 into OpenAI—avoiding the current manual screenshotting and scraping process. Hope springs eternal in the bava breast.
Monday Night Football featuring Colts vs Jags which should be at the RCA Dome, not Lucas field—although I still love the graphic
In terms of new developments since last week, one of the things I ruminated about in the “Stats and Intangibles” post was how I might be able to have game previews and recaps written about the other 13 games each week that are simulated by Madden 2001. There are no game play specifics, rather just the team and individual stats before and after the each week’s game.* So my idea was screen shot all the stats for each team before the match-up (including their record) and have ChatGPT provide a compelling preview of the coming week’s game. Here are two previews ChatGPT created for the Falcons vs Rams and the Colts vs Jaguars. I have to say, the breakdown of the Colts strengths gives me a lot to think about given they are divisional rivals:
The Colts come in unbeaten at 2–0 after wins over the Chiefs and Raiders. Peyton Manning has started steady, throwing for 379 yards and 2 TDs, and the offense runs through Edgerrin James, who has already piled up 300 rushing yards and 4 TDs at 5.4 yards per carry. Indy’s defense has been sharp, giving up just 31 points across two games, and ball security has also been a strength: just 2 giveaways so far.
It kind of feels like scouting other teams as the Jets prepare for future games. But the real fun part was having ChatGPT hallucinate game highlights. For example, I loved the recap of the nail biter in Atlanta that saw the Rams eek out a win against a tough Falcons team:
A Wild Fourth Quarter
The Rams continued to chip away at the lead early in the fourth with a Jeff Wilkins 32-yard field goal to make it 17–16. But Atlanta’s kicking game was sharp too: Jay Feely drilled a 38-yarder, then added another from 36 yards to stretch the Falcons’ lead to 23–16 with just under two minutes left.
That left Warner with the ball — and the game in his hands. Starting from his own 25, he strung together completions to Torry Holt and Ricky Proehl, methodically moving the Rams across midfield. With the final seconds ticking away and Atlanta dropping deep to protect against the field goal, Warner pump-faked to Holt and found Bruce streaking on a skinny post. Bruce hauled it in at the 10, cut inside a diving defender, and scored the go-ahead touchdown with 3 seconds left on the clock.
It’s like the game happened in some alternate universe wherein ChatGPT helped streamline the imaginary. So, my experiment with using pre and post-game stats each week for simulated games worked. The other piece would be having some AI-engine create GIFs and video clips of the imagined moments for the highlight reels. But for right now I’m pretty happy with the outcome. it underscores that it would be doable to turn AI Maddeness into a full blown league news preview and recap site, rather than just a myopic focus on the Jets. I want to see how much writing previews for all the games adds to the weekly time sink, guessing it’s significant with my current manual methods.
Anyway, things are moving along. After just a few weeks in it’s become readily apparent how folks get sucked into the machine. I’m spending too much time solipsistically writing into a chat box, which can feel really alienating. That said, there are also moments of joy when I’m endlessly entertained by some of the insane results, like this cartoon of Curtis Martin steam rolling the Tampa Bay Buccaneers:
Curtis Martin steam rolling the Tampa Bay Buccaneers defense
While I’m enjoying my AI Maddeness project, I’m also thankful it has a definitive end date just a few months from now. It can quickly consume my week if I’m not careful and the last thing I need right now is another hobby to feed my unbridled mania.
That said, what would really make AI Maddeness amazing is if you could have 31 people (or however many teams there are that year) each take ownership of a team and you have weekly games on Sunday that are played and streamed live. After that, each of the teams manage their own PR and marketing but the AI Maddeness admin (the league commissioner) manages each week’s previews and recaps in an attempt to remain somewhat unbiased while the various teams drive their own media empire based on the season. Each participant gets a site for their team, another for a local news outlet, and a fan site to start. For this setup to work we would’ve had to already figured out pulling in targeted team stats into each of the sites seamlessly, and the recaps and previews could be automated based on those stats and any specific parameters each person in-charge of the franchise wants to customize. It’s something we have time to work up to given Madden doesn’t get online head-to-head play until 2003.
___________________________________
*Turns out I was wrong in that post when i said madden 2001 has an incomplete record of individual player stats on display. All the stats are there for each and every player, I just didn’t do the quick gamepad switches to see them. The have full stats for all players, with everything from leaders in the league, conference, division, and team. It’s all there, so no more excuses, Jimmy.
