Do you know what time it is?

I said, DO YOU KNOW WHAT TIME IT IS, PEOPLE??? IT’S TIME FOR THE NFL DRAFT, SO LET’S KICK OUT THE JAMS … and watch Mel Kiper Jr. shake his moneymaker for 72 consecutive hours before metastasizing into Larry King.

(Actually, if all NFL players had hair like Kiper’s, with its M&M-like outer shell covering the chocolate-flavored goodness inside, concussions would cease to exist. And if John Glenn had had a Kiper instead of that dorky Marine ‘do he could have taken Friendship 7 to freakin’ Mars.)

I love the NFL Draft – not for Day 1, which is the real Longest Day, or for Day 2, when the hologram of Bronko Nagurski announces all the picks, but for Day 3, and specifically the end of Day 3, when Adenoid Adrenalopathy from Piscataway Tech finally comes off the Big Board.

You can keep me cryogenically frozen for 364 days out of the year, but thaw me out for that.

One of the things I do obsessively in advance of the real NFL draft is conduct pretend NFL drafts. They’re called “mock drafts,” after what you’re allowed to do after seeing me do 18 of them every stinking night.

The computer-powered platforms that power these mock drafts are not terribly technologically advanced. ESPN can make one, for crying out loud.

The computer makes its picks based on what it perceives to be team needs, makes a few trades, and then stops when it’s time for you to pick.

You pick, it chortles to itself, and then the wheel spins again until it’s time for your next picks.

Some mock-draft tools include a grading mechanism at the end, where you are awarded a D+ for picking Kool-Aid McKinistry at the top of the second round, though every draft “expert” worth his Aqua Net swears that McKinistry is going mid-to late-round first.

Given that mock-drafting is a relatively linear process with a large amount of training data (since there have only been about 2.1 million mock drafts published since Christmas), you’d think that AI would do pretty well at mock-drafting.

After all, AI is supposedly great at predicting what comes next based on the training data, and it’s good at following rules, to wit:

Caleb Williams is No. 1, and then there’s a 52% chance it’s Jayden Daniels at No. 2.
“Caleb Williams” and “USC” and “QB” always go together, in that order.
Chicago picks first, then Washington, New England and Arizona, unless someone makes a trade.

Given the fact that AI can tell nuclear physicists how to master cold fusion, you wouldn’t think this is beyond the capabilities of AI. Only AI cannot do it. It is objectively horrible at drafting in ways that I'm barely able to comprehend.

If I ask ChatGPT to “conduct a 2024 NFL mock draft,” these are the results:

Outside of the Panthers being thrilled they got C.J. Stroud in the redo, not a lot went right here.

(Extra love to the comment after the Texans drafting Drake Maye: “The Texans might look for a quarterback to lead their rebuild.” They might. And his name might be C.J. Stroud.)

You know, maybe it’s a ChatGPT thing. So I asked H2O.ai the same question. The cool thing about H2O.ai is that it uses four different AI engines, so it gives me four times the misinformation for the price of one. Such a boggin!

Caleb Williams or C.J. Stroud? I don’t know how the Texans were able to choose.

Finally I asked perplexity.ai, the so-called “answer engine.” And it had an answer for me all right:

Caleb Williams is such a great prospect he was drafted twice! And none of the trades appear to be with anybody for anything. They’re just trades. Finally, it’s unclear how the Texans wrangled the No. 2 pick. Good behavior, perhaps.

Note that these are relatively good results. One iteration had the Carolina Panthers drafting Memphis Grizzlies point guard Ja Morant, which outside of spelling “Ja” correctly is wrong in every way it’s possible to be wrong.

(Update: I did another iteration with Perplexity – yes, I’m still doing these – where Zay Flowers was drafted 10 times in the first two rounds. He truly is Everybody’s All-American. Or, as Chuck Klosterman wrote in the greatest music review ever, “How awesome it must be to be awesome!”)

You can see the hierarchy of errors here. Basically, ChatGPT is:

Not adhering to draft order
Not adhering to player/position/school rules
Not following “best practices” for prospect hierarchy as established in thousands of prior mock drafts
Not adhering strictly to the pool of players eligible for the 2024 NFL draft

So maybe that’s it. Maybe the problem isn’t that AI sucks at this but that I suck at telling AI how to organize and present this information.

This is entirely possible, so I tried again with a more advanced prompt, with more instructions and more guardrails.

And … nothing really changed. First, ChatGPT:

Then Google Gemini:

(“The actual draft could unfold differently”? Damn straight it could.)

Then H2O.ai:

(Note to H2O.ai: Tyler Van Dyke is still in school.)

And finally perplexity.ai:

Saying this is the best of the bunch only points out how bad the rest are.

What’s going on here? The training data could be old. Depending on when the appropriate data was scraped there could have been more information on 2023 prospects than 2024 prospects, even though we specifically requested that AI draw from the 2024 prospect pool.

It could be that the data wasn’t specialized enough. It’s plausible that the AI tools would do at least as well as the mock-draft simulators if they had the same database to work from – though similar databases are all over the internet.

Also, we might not have laid out the rules in an AI-appropriate manner, though these tools are designed to work with suboptimal prompts.

Ultimately we have to accept that mock drafts are one of the many things AI does badly, along with learning aids for children and 3-D magic pictures.

Janelle Shane, the genius behind the AI Weirdness website, likes to point out that people are trusting their lives and livelihoods to a tool that doesn’t know its basic shapes and calls a frog a chicken.

I’ll take that one step further: How in the name of all that’s meaningful to this great nation can we let a so-called artificial-intelligence tool make decisions for us when it blasphemes all that is holy by placing Jayden Daniels on the University of Utah?

I expect Senate hearings within the hour.

The AI Is Objectively Bad At Mock Drafts