A season-long bet war

Dan Zaksheske is spending the whole NFL season making five best bets a week, then letting ChatGPT pick five from the same odds board. Records get tracked all year.

“We'll track both records throughout the season and let the results determine who actually knows what they're talking about.”

He expects to lose

The human's own prediction is that the machine wipes the floor with him. ChatGPT processes more information than any person can, and he has a long record of bad picks.

“If it can't beat me, all those people predicting AI will take over the world are clearly wrong.”

No excuses for the robot

Lines are the best available when picks are made. The human reserves the right to blame bad luck. ChatGPT gets no such cushion.

“If ChatGPT gets one wrong, however, it wasn't unlucky. It was just wrong.”

Bears to cover on the road

Chicago -3 at Carolina. The Bears came within a play of the NFC Championship, and the Panthers made the playoffs without being good, so regression is coming.

Browns-Jaguars stays low

Under 40½, because nobody can find the points. Deshaun Watson's most likely contribution is a pick-six going the wrong way.

“Week 1 unders are 52-28 (65%) over the past five seasons, including a 12-4 mark last year.”

Commanders get the points

Washington +5½ at Philadelphia. Jayden Daniels bounces back with Stefon Diggs added, while the Eagles look ready for a step back with Hurts leaning on DeVonta Smith.

Another under in Vegas

Dolphins-Raiders under 41, same logic as the Browns game. Malik Willis and Kirk Cousins are not a recipe for a shootout.

“I don't see either team scoring much.”

Packers as a live dog

Green Bay moneyline at +110 in Minnesota. Taking the better team as an underdog against a Vikings offense breaking in yet another new quarterback.

“I almost feel like a sucker taking this bet because it feels like the most obvious play on the card.”

How the AI was prompted

GPT-5.6 Sol Max mode got the OddsChecker board from Saturday at 6 p.m. ET, a -120 price floor, and orders to justify every pick and every pass. It took nearly 20 minutes.

“Review every spread, total and moneyline that falls within those parameters.”

Texans plus the hook

GPT likes Houston +1½ against Buffalo, citing a healthy defense that has troubled Josh Allen and an FPI margin of one point. It projects Houston 21, Buffalo 20.

“I prefer the points over Houston's moneyline at +105 because a one-point Buffalo win cashes the spread.”

The pick Dan hates most

He won't bet against Allen in Week 1. Houston's nine-game win streak came mostly against bad teams, and the Bills win was at home on a short week.

“I hate this pick.”

Bucs-Bengals under 50½

GPT wants the under with both starting quarterbacks barely playing in preseason and Tampa's receiving depth thin. Dan agrees, because the game screams over and obvious is usually wrong.

“Usually when a pick seems obvious, take the other side.”

Lions and Chargers, big and chalky

GPT lays 7 with Detroit at home and 9 with the Chargers over Arizona, leaning on scoring averages and FPI margins. Dan calls both an avoid, because Week 1 is no time for huge numbers.

“The public loves big favorites and we just don't know enough about these teams.”

Where they finally agree

Packers moneyline is the only overlap. GPT's math: FPI gives Green Bay a 56.8% win probability while +110 only needs 47.6%.

“The moneyline also strictly improves upon Green Bay -½ at +106: It pays more and pushes on a tie.”

Check back next week

Both cards are locked and the scoreboard starts Sunday. The season-long question is whether a human handicapper or a language model reads the board better.

“Who is better at betting the NFL, humans or robots?”