Wednesday ORT: Miami GP, MDG's Flawless Victory, AI Hallucinates, inEVitable Defeat

All subscribers welcome

I’m in Germany for the second half of the week so don’t be surprised if I respond to your comments while you’re asleep. Depending on my ability to find a power adapter, we might have an extra car review Friday… if not, it will be Saturday.

“You want me to let him pass as well?”

When it comes to Formula 1, these are the good old days. Will the scrap between Oscar Piastri and Max Verstappen in the opening laps of the Miami Grand Prix eventually be seen as the crucial moment in the drivers’ championship? If nothing else, it showed that Piastri is completely unafraid of the current WDC title holder. While I don’t think Max played dirty, he certainly raced hard. It didn’t discourage Oscar, who pressed on regardless and eventually took the position. In fact, the way the McLaren driver pushed hard from fourth place to the win was, ahem, refreshingly Verstappen-esque, and a far cry from the way Lando Norris has traditionally performed when starting from pole. Great race, and surely a thrill to watch for everyone but Aubrey “Drake” Graham, whose $20 million bet at the beginning of the season would have been much better placed on the #81 than on the #4.

Other notes:

• Kimi Antonelli is obviously the real deal. How satisfied must Toto Wolff be, having lost Sir Hamilton’s salary while gaining a faster driver?

• Also the real deal: Cryin’ Hadjar, who is whipping Liam Lawson almost as badly as Max did.

• Good riddance to Mick Doohan. Helmet Marko was right: he’s a third-tier driver and his decision to berate the team after losing what has to be considered a very slight chance to escape Q1 does him no credit whatsoever.

• The Lewis/Leclerc radio drama was unnecessary and speaks to Ferrari’s general inability to execute strategy. If they’d let Lewis through immediately, he’d have caught Kimi. But if they’d never let him through, Charles would have caught Kimi. It took some real effort to make sure neither car got up there while at the same time encouraging Carlos Sainz to hunt down his old team and make a real statement of a move on the fellow who took his job.

• Time will tell if a seventy-five-year-old Flavio Briatore can further improve Alpine’s fortunes, but you would be foolish to bet against him.

F1 Watch watch: Tudor celebrated the Miami Grand Prix by introducing a Racing Bulls limited edition chronograph. It’s carbon fiber and costs $7,575. And I thought the $5,100 Black Bay Ceramic was ambitiously priced! Here’s the funny part: like the turqoise dial chrono released last year, this already-expensive watch will probably fetch twice the original MSRP in the grey market. Have we really gotten to the point where we have to start buttering up our Tudor authorized dealer? What’s next? Buying three Tissot quartz watches to get a PRX Carbon allocation? Fear not, Tudor fans: you can get the same case construction and movement in the Pelagos FXD Chrono for two grand less, immediately.

Last to first, in the rain, past the occasional vice president

This was the first race weekend of the year for Green Baron Motorsports. Mini Danger Girl and I ran the club event at Waterford Hills, a track we both like, while MDG’s mom stayed at home to work on her race car.

MDG missed Friday practice due to mechanical issues and I skipped said practice because I don’t particularly need it. Come Saturday, which was dry, I qualified in second place, well behind the lightning-quick Formula Continental of Sean Gennari. Unfortunately for Sean, he had a contact incident with his own brother, who also runs an FC, on the final lap of qualifying. Both cars were seriously damaged.

I went flag-to-flag on this one despite having very little tire left after qualifying. The second and third place cars were both faster than I was, but my start was strong enough to give me breathing room and I finished a comfortable 0.4 seconds ahead. My final lap was a 1:07.998, which is six-tenths ahead of my previous best in the PR6 and very close to the all-time Radical lap record of 1:07.528 set last year by J. Baruth in a Radical SR8. It says quite a bit about Waterford Hills that a 165hp car is basically dead even with a 445hp car.

On Sunday I parked the Radical because both of our cars had an electrical issue and I wanted to focus on MDG’s race. She had 16 cars in her group and had qualified next to last, but there was severe weather coming in. A tour of the paddock showed that most of the competition was on slicks, betting that the rain would stay minimal. I was of the opinion that we would gain nothing by following everyone else’s strategy, so I put MDG on rain tires a full 20 minutes before grid, in the hopes of catching major weather during the race and moving up the order as a result.

During those 20 minutes, the weather got worse, but some drivers refused to change tires. At the five minute mark, it was obvious that this would be a full rain race. Everyone rushed to make the change. Not all of them got it done. I audited the grid, told MDG which cars would be vulnerable, and told her that it was her time to shine.

Which she did, making up five positions in the first lap. This is only her fifth race weekend, and the first one in these conditions, but she has a natural touch for traction seeking and in her video you can hear how disciplined she is on the throttle. She ran down the leaders with icy consistency and finished eleven seconds ahead, with a fastest lap almost 2.5 seconds better than that of the next best driver.

Most of her fellow competitors, including my friend and former co-worker Alex Della Torre, were thrilled for her. It was a bit of a fairytale story, really. She is still just 21 years old. Imagine how much better she can get from here. I can’t wait to find out.

EV, BTFO

well this all sounds perfectly ducky

Whether you consider Donald Trump to be the God Emperor or a washed-up reality-TV star, you have to admit that in his second term he is certainly doing a better job of keeping the promises that got him elected. The latest example: revoking Biden-era EPA directives that were intended to get the country to 50% EV car sales by 2030. His administration will also seek to enjoin California and its fellow-traveler states from mandating 100% EV new-car sales by 2035.

This is good news for everyone, including EV fans. I’ve always found it fascinating how the same people who decry the anticompetitive nature of tariffs will turn around and espouse the worst sort of government manipulation in the automotive market. (Feel free to reverse those two examples, my blue friends; the ironing is equally delicious on either side.) The massive tax subsidies and perceived inEVitability of “the electric future” have done tremendous damage to EV sales in the United States.

A good electric vehicle won’t need massive subsidies, nor will it need to be sold at gunpoint. I’d be remiss if I didn’t mention that the oil and gas industry is also subsidized by any and all means up to and including regime change, but oil is a fungible commodity and in the long run the price you pay for gasoline isn’t far removed from what it costs to extract it. EVs are already subsidized after a similar fashion by public utilities. There’s no need to subsidize their purchase further.

Were I the “car czar”, I’d start leaning heavily on automakers to reduce weight and increase fuel efficiency by conventional means. Here’s a wacky idea: have a maximum height for vehicles on personal-use plates, and make it about as high as a Rolls-Royce Phantom. That would put another 10mpg on the fleet economy of new vehicles, for sure. If you want something taller than that, pay a tax — and maybe that could subsidize appropriately-sized EVs.

In the meantime, I think President Trump acted correctly. The 50% future of 2030 was going to be a supply-and-demand nightmare. You’d see new gas-powered cars fetching ten grand over MSRP. It would make the Covid-car era look rational. I’m glad it won’t happen.

The only “hallucination” is taking place in the heads of AI “experts”

Last month, an A.I. bot that handles tech support for Cursor, an up-and-coming tool for computer programmers, alerted several customers about a change in company policy. It said they were no longer allowed to use Cursor on more than just one computer.

In angry posts to internet message boards, the customers complained. Some canceled their Cursor accounts. And some got even angrier when they realized what had happened: The A.I. bot had announced a policy change that did not exist.

So says the Times in one of several mainstream media stories appearing this week about “AI hallucinations”.

The newest and most powerful technologies — so-called reasoning systems from companies like OpenAI, Google and the Chinese start-up DeepSeek — are generating more errors, not fewer. As their math skills have notably improved, their handle on facts has gotten shakier. It is not entirely clear why.

It’s entirely clear why, folks. Even to me, someone who has done just a tiny bit of machine learning programming in his career. (Some of you know that I currently earn my daily bread in the “AI” space, by the way; I’m advocating the truth here over my own economic and career interest.)

The early “AI” results were impressive in large part because they were deliberately based on a fixed, and aged, set of data. ChatGPT and its cousins deal in “token prediction”, which is a fancy way of saying that they give you an approximation or averaging of what they’ve ingested. When you work with old data that isn’t changing, this works pretty well. It helps if you filter and adjust the data a bit, as all the early AI providers did.

In 2025, however, we allow “AI” to work in the present day with data that, in many cases, has already been corrupted by AI token prediction. Nor are we particularly good at telling AI what to ignore when it predicts tokens.

Let’s make the cardinal error of thinking that Cursor’s AI bot is a person. Imagine this person hearing about Netflix and Spotify and other major providers limiting their product to one computer at a time. Imagine him seeing the general trend of limiting subscription use. All of this increases his general “feeling” that products should only be used on one computer. And unless someone tells him specifically that Cursor can be used on more than one computer, he will eventually “guess” that Cursor is pretty much like all the other products, and he will tell his clients about the new restriction.

The critical thing I want you to understand about “hallucination” is that it is built into large language models, which have no genuine sense of truth or reality. The only difference between a “regular result” and a “hallucination” is whether you, the human being interacting with the AI, think the result is true or not.

Want some proof? Ask your LLM of choice to complete the phrase

The first cut is the —

You know what it’s going to say, right? The first cut is the deepest. Because that’s a song whose lyrics have been spammed all over the Internet. It’s the most common, and therefore the most predictable, token.

But it’s not true in any sense of the word. There is nothing to say that the first cut would be the deepest.

Now imagine this large language model in charge of doing emergency-room triage.

Hint: if you want to get seen in a hurry, try telling the ER kiosk that… “I’ve got the first cut!"“