Draft Day or ChatGPT on drugs
Draft day (or draft weekend) is the best time of the year — next to my wedding anniversary, my kids’ birthdays, and all family holidays (just covering all my bases here).
Our league has evolved from a modest 10-team snake draft, crowded at a table in Pluckers in Austin, Texas, to a 14-team league with an ever-growing list of managers and friends. Our league is 17 years old and we’ve drafted across the country, from a memorable 10th anniversary in a Vegas suite to renting an entire camp with cabins, a pool, and a game room (the draft on a cruise ship is still on the list).
These are the people I’ve grown up with — the ones I can be completely myself around. But when that auction draft starts, it’s all business.
While our preparation has waned over the years, even our ‘unprepared’ managers would out-prepare most other leagues. We know every draft strategy, from zero RB to late round QB, and spend hours trying to outguess each other’s psychology. Yet, we’re always wrong.
This year, I went all-in with AI, specifically ChatGPT. I fed it everything — our scoring rules, lineup positions, manager names, keeper players, and draft trends. ChaseGPT (TM) gave me a high-upside roster to target and promised to adjust in real-time during the draft.
But then, ChaseGPT fumbled at the 1-yard line.
At a biergarten with laggy Wi-Fi, I set up my laptop and started the draft. ChaseGPT provided real-time suggestions — until about round three, when a pop-up informed me I had overloaded the chat. Starting a new conversation caused ChaseGPT to lose its footing. It began hallucinating, giving inconsistent suggestions, poorly formatted responses, and failing to track my budget properly.
This was the beginning of the end. My preparation was apparently too much for my one chat and I needed to start a new conversation. Unfortunately when I started my new conversation ChaseGPT had trouble remembering how we were interacting before and began to hallucinate. I was getting inconsistent suggestions, poorly formatted responses, and it was no longer keeping track of my budget properly.
It was all downhill from there.
Despite targeting most of the players I wanted, they went for higher prices than planned. Worse, when one of my targets was bid up, I was left scrambling. By the end of the draft, I had one elite player and a bunch of guys with upside.
Post-draft, it got worse. When I asked ChaseGPT about Aaron Jones, it hallucinated and told me he was still on the Packers, even though I had repeatedly told it he was traded to the Vikings. The same happened with D’Andre Swift.
My final roster was underwhelming — probably the worst in the league — and heavily reliant on the success of the Bucs and Bears (not ideal).
So here’s what I’ve learned: ChatGPT can give the impression that it’s up to speed and ready to help you draft strong, but if it errors or hallucinates, you could be f***ed. It struggles to adapt to irrational drafting decisions by managers and might stick to its guns when a human would switch strategies. While ChatGPT 4o has access to current information, unless prompted, it may rely on outdated data from August 2023.
Had left ChaseGPT to its own devices, my roster might have been even worse. But maybe it knows something I don’t? Now, it’s time for in-season management…
