Photo: Bill Streicher, USA Today Sports

Gambling on the NFL is big business, especially after a 2018 Supreme Court decision striking down a federal ban on sports betting. Recent estimates suggest that as many as 46.6 million people will place a bet on the NFL this year, representing nearly one out of every five Americans of legal gambling age. As a result, there's been an explosion in sports betting content, most of which promises to make you a more profitable bettor. Given that backdrop, it can be hard to know who to trust.

Fortunately, you can trust me when I promise that I'm not going to make you a more profitable sports bettor. And neither will any of those other columns. It's essentially impossible for any written column to do so for a number of reasons I'll detail over the year. (I'm not saying it's impossible to be profitable in the long term betting on the NFL, just that it's impossible to get there thanks to a weekly picks column.)

This column's animating philosophy is not to make betting more profitable but to make betting more entertaining. And maybe along the way, we can make it a bit less unprofitable in the process, discussing how to find bets where the house's edge is smaller, how to manage your bankroll, and how to dramatically increase your return on investment in any family or office pick pools (because Dave in HR and Sarah in accounting are much softer marks than Caesar's and MGM).

If that sounds interesting to you, feel free to join me as we discuss the weekly Odds and Ends.

Checking In On the Unders

In Week 7, I noted that unders had been hugely profitable so far this season and discussed structural reasons why unders tend to outperform overs (though usually not by enough to beat the vig). I also said I'd track the performance of the unders going forward to see if we could be profitable merely by mass-betting them every week. (My hypothesis was that unders would win somewhere from 50-52% of the time going forward.)

Unders went 6-8 in Week 10, costing us 18.2% of our wagers and giving us our first true losing week (though unders were 8-8 in Week 8, and we lost money on the vig). Overall, unders have gone 31-24-2 since we started tracking; if you bet an equal amount on every game (and all action was at -110), you would have turned a 7.3% profit. At $10 per bet, your total profit to date would be $41.82. If you'd instead started with $160 and invested an equal percentage in every bet every week (rolling over your winnings or losses), you'd be up $38.13, or 24% of your starting bankroll.

Gambling With AI

Part of this column's schtick is that I do the things that all the other betting touts do; I'm just transparent about it. I use selective endpoints and sleight of hand to frame my record as favorably as possible (assuming I don't just lie about it outright). I take advantage of line movement to grade myself against betting opportunities that might no longer be available by the time my readers placed their bets.

I give picks at random and then manufacture a justification for them after the fact. I create multiple "products" ("Lock of the Week", "Rematch of the Week", "Revenge Game of the Week") to maximize the chances one is successful, and I can focus attention on that while distracting attention from the unsuccessful predictions. I wrote an entire column on how to leverage selection bias to guarantee that you create a pool of readers who see a wildly disproportionate amount of successful picks.

The problem with the industry is one of misaligned incentives. The things that will do the most to help readers become more profitable bettors (or more realistically, to become less unprofitable bettors) are very different from the things that will do the most to help writers become more profitable content creators. "Maybe mass-bet unders?" is a neat trick, but it's the sort of thing you can only write once, which isn't great when trying to build a recurring revenue stream.

To stick with the schtick, the rest of this column will be written by AI. Or rather it will be rewritten by AI, as I feed the first draft into ChatGPT and ask it to edit and rewrite in the style of Hunter S. Thompson.

This Column Was Written By AI

In the twisted carnival of digital minds, Artificial Intelligence is a creature of many faces. It's a slippery bastard, and just as crucial as knowing what it is, is grasping what the hell it ain't. Take Supervised Artificial Intelligence, for example. It's got "supervised" in its name, but it's neither watched over, nor genuinely artificial, nor remotely intelligent. It's like calling a toothless hound a fearsome guardian.

Large Language Models (or LLMs, if you fancy acronyms) are a peculiar breed. They're like having a blurry snapshot of the entire damn internet, smeared with the eccentricities of human thought. Someone once described them as "plausible sentence generators," which is a polite way of saying they're skilled bullshitters. Sure, accuracy ain't their strong suit – these LLMs can spin tales, hallucinate like a psychedelic trip, and swear it's gospel truth. But, hey, they're built to sound damn convincing. Ask them to write a betting column in the style of Hunter S. Thompson, and by the twisted gods of journalism, you'll get a piece that smells like it came straight from the wild mind of a gonzo scribe or at least a second-rate imitation.

Now, here's where the carnival barker gets excited. If you want plausible sentences, these LLMs are the dark sorcerers of the text realm. Stick a human in the loop to keep these electronic scribes in check, and they're proofreading wizards. They can mimic the scribblings of a disinterested high schooler, a pompous professor, or even the ghost of gonzo journalism's father himself.

But accuracy? Ah, there's the rub. Accuracy is for the weak-hearted, my friend. If you're all in for plausibility, these LLMs will crank out miles of sentences faster than a bat out of hell, and you won't even break a sweat. In the seedy world of gambling advice, where plausibility is more precious than a sober moment, this is gold. Picture this: a swarm of content flooding the arena, crafted by invisible hands and cost next to nothing for the puppet master behind the curtain. Since all gambling advice is a coin flip, whether it's spawned by an "AI" or a human hand, it's a 50/50 gamble for the poor soul on the receiving end.

And let's not forget the marketing razzle-dazzle. Slap an "AI" label on your advice, and suddenly, you're the oracle of the technofuture, extracting gold from the pockets of unsuspecting marks. Sure, the so-called Public-facing LLMs were trained on ancient scrolls of data, oblivious to the current NFL scene. And yeah, machine learning can predict football games, but the bookies knew that trick before you even dreamt of it, and their crystal ball is probably shinier than yours.

In the sordid landscape of gambling advice, AI-generated content is the new player in the rigged game. Odds and Ends is hitching a ride on this twisted rollercoaster, dipping our toes into the AI-generated chaos. But here's the kicker: our content may dance with the devil of AI (for this week only, mind you), but our beating heart remains the same. The only AI we'll ever trust is our old pal, the pseudorandom number generator – just as accurate as AI and as real as the smoke and mirrors of this digital sideshow.

Lines I'm Seeing