The vibes and the noise

The post-empirical generation.

A programming note: Have you ever thought, “These blog posts are alright, but I wish that they were longer and louder? Well. This post was adapted from a recent talk, so if you’re sick of mere metaphorical yelling and would prefer actual yelling, there is a video of that on YouTube. Like and subscribe.

---

Here is what happened to Jordan Chiles:

1. In 2024, Chiles qualified for the Olympic women’s floor exercise final. In the final, every competitor performs one routine, which receives two scores: A starting difficulty score, which is determined by the elements in the routine, and an execution score, which is awarded by a panel of judges and added to the difficulty score. Judges also impose standardized deductions for penalties, like falling or stepping out of bounds, which are subtracted from the execution score. All of this is carefully documented and diagramed in the 214-page Code of Points.

2. Chiles was one of nine qualifiers in the final, and was scheduled to perform last. After the first eight gymnasts performed, Rebeca Andrade, a Brazilian, was in first place with a score of 14.166. Simon Biles, who stepped out of bounds twice, was 0.033 points behind Andrade. Then, two Romanians—Ana Bărbosu and Sabrina Maneca-Voinea—were tied for third with scores of 13.700. Bărbosu held the tiebreaker and was in position to win the bronze medal.

3. Chiles did her routine.

4. She got a 13.666, scoring a 7.866 on a routine with a difficulty score of 5.8. So, fifth place; no medal.

5. But! Chiles’ coach noticed that her difficulty score was calculated incorrectly. It should’ve been a 5.9—which would’ve made her final score a 13.766, moved her ahead of the Romanians, and put her in third. The coach protested; the protest was upheld; Chiles was moved to third place; a bronze medal; USA, USA, USA.

6. But! Bărbosu’s coach noticed that Chiles’ coach took too long to protest Chiles’ score. According to Article 8.5 of the 162-page 2024 Technical Regulations, “inquiries for the Difficulty score are allowed, provided that they are made verbally immediately after the publication of the score or at the very latest before the score of the following gymnast/athlete or group is shown.” But, because Chiles went last, no gymnast followed her. So, “for the last gymnast or group of a rotation, this limit is one (1) minute after the score is shown on the scoreboard,” and “late verbal inquiries will be rejected.” Bărbosu’s team reviewed the tape, said that Chiles’ coach filed her protest after one minute and 24 seconds, and that her protest should be disallowed.

7. But! Maneca-Voinea—the other Romanian—saw that her score included a 0.1 point deduction for stepping out of bounds. So she also reviewed the tape, and saw that she did not, in fact, step out of bounds.

8. So she protested, saying she should not have received the deduction, and that her score should’ve been a 13.800, which is higher than Bărbosu’s score and both of Chiles’ potential scores. So she should be in third place; a bronze medal; ROM, ROM, ROM.

9. The Court of Arbitration for Sport—a legal body that arbitrates disputes in international sports¹—investigated the challenges. It found that “the decision as to whether a 0.1 deduction was appropriate is a textbook example of a ‘field of play’ decision, one that does not permit the arbitrators to substitute their views for that of the referee,” and “it cannot be reviewed.” In the case of Ms. Maneca-Voinea, “the challenge is dismissed.”

10. In the case of Ms. Chiles, they watched the video footage. They consulted an “official report prepared by Omega, the official timekeeper for the Olympic Games.” They interviewed Ms. Donatella Sacchi, President of the Women’s Artistic Gymnastics Technical Committee within the Fédération Internationale de Gymnastique, or FIG. And they concluded that “in the view of the Panel, the words ‘one minute’ in Article 8.5 mean one minute, no more and no less,” and the Chiles’ “inquiry was made within one minute and 4 seconds.”² Therefore, “Ms. Chiles’ inquiry is to be accordingly dismissed and her initial score of 13.666 reinstated.”

11. So, all the scores were reset to their original values; Chiles was moved back into fifth; Maneca-Voinea stayed in fourth; Bărbosu was moved back to third; a bronze medal; ROM, ROM, ROM.

12. Then, chaos. The Romanian prime minister boycotted the closing ceremonies; Romanian great Nadia Comăneci took it to the streets; people fought in her comments; Chiles sued again;³ the U.S. Olympic and Paralympic Committee accused the arbitrator of having ties to the Romanian government; other lawyers proposed punting on the whole mess, and that everyone should tie for third; three bronze medals; ROM, ROM, USA.

13. It has been 494 days since the final, in which three women each performed for a combined 270 seconds, and, as of three weeks ago, “the Swiss Federal Supreme Court has not issued decisions on either the setting-aside application or the request for revision.”

At the beginning of Section I of the Code of Points, the document declares its purpose: “To provide an objective means of evaluating gymnastics exercises.” Section II defines the rights of the gymnast in this evaluation; the very first right—article 2.1.1a—is the right to have performances judged correctly and fairly.

Which demands the obvious question: Despite the 400 pages of rigor and quantified rules, was this competition judged correctly? Was it objective? For all three women involved, does anything about this result seem fair?

And how did all of this even happen?

The quantification of everything

You could answer that question in two ways. The proximate cause of last year’s mess was another, similar mess in 2004, in which a scoring error—also about an incorrectly assessed deduction—cost Yang Tae Young, a South Korean gymnast, the gold medal in men’s Olympic all-around. FIG overhauled the scoring system after that, replacing the traditional “perfect ten” framework with the new, more mathematical Code of Points.

But the Code of Points could also be explained as part of a larger narrative: The quantification of everything. Because, in the late 2000s and early 2010s, numbers became our inexorable future:

“It’s a revolution,” says Gary King, director of Harvard’s Institute for Quantitative Social Science. “We’re really just getting under way. But the march of quantification, made possible by enormous new sources of data, will sweep through academia, business and government. There is no area that is going to be untouched.”

Hedge fund managers no longer bet on their intuition, but on their models. Sports teams did the same. Political pundits were out; Nate Silver was in. He launched FiveThirtyEight as an entire media division, to bring their “data-driven approach into new areas.” We quantified companies, elections, and ourselves. And, gymnastics.

As a sub-trend: We also wrote blog posts about everything that we could graph: Our sleep; our commutes; our mortality. Writing in the New York Times, Seth Stephens-Davidowitz looked at the numbers behind our musical preferences. He found that our tastes are overwhelmingly defined by what was popular when we were teenagers. More recently, the Washington Post’s “Department of Data” extended the idea, and found that the same pattern holds for nearly everything: Fashion, movies, television, sporting events, food—all of it, we said, was better in our adolescence.

Of course, there is no “best fashion.” There are only fads, and the generational indoctrination that makes us believe that our fad was the best one. With things like clothes, no matter how militantly we believe that we looked cool and that the Kids These Days look dumb, it is easy to see how fickle fashion can be. It is easy to see that, as our styles fade into obsolescence, the next generation is not wrong; we are just growing old.

But other trends can be harder to recognize as trends. For example: What makes a good employee? What are the right ways to think, and make decisions? What is the best way to answer an ambiguous question like which of these products, or baseball players, or gymnastics routines, is best?

If you came of professional age in the 2000s and 2010s, your answer—per the same trend that created Stephens-Davidowitz’s research—is to be data-driven. But is that answer right—is it the “best method”?—or do we simply believe that is, because that was the corporate philosophy that we were indoctrinated into? Was the quantification of social science—and business, and culture, and sport—a revolution, or was it just a fad too?

The new revolutionaries

When this all started, 15 years ago, there was clout in data work. It was urgent; it was prestigious; it was strategic; it was, dare we say it, cool:

Sterrett saw analytics as “pre-eminently the profession of business advice” and the analyst as a person who “is thoroughly conversant with the principles” underlying a successful company and who “has accumulated a large fund of information in…business policy.”

And:

Stettler believed that the prestige of the data profession would grow to match and surpass that of the older, more recognized professions of law and medicine, and that data scientists would outnumber physicians and lawyers.

But cool never lasts. Because those two quotes aren’t actually about analytics and data science; they are about accounting, from 1904 and 1968. And accountants, critical as they are, are rarely in charge these days. They are fact-checkers, not decision-makers. They are the part of the corporate machine that someone else drives.

For analytics, the path seems similar. A generation grew up reading The Signal and the Noise and Thinking Fast and Slow, and bludgeoned the prior generation to death with models and quantitative rhetoric and complex codes of points. In God we trusted; all others must bring data.

No longer. Anu Atluru is right: Taste is eating Silicon Valley. Craft, not data, is the new buzzword. Linear CEO Karri Saarinen, whose company has built one most copied brands of the 2020s, recommended that startups “ban use of data as a decision making tool.” “Don’t make decisions based on data or experiments,” he told Figma; “to design with craft, you must develop and trust your intuition.” The former dean of Harvard Business School—whose associated publication, the Harvard Business Review, has published hundreds of articles on the urgency of becoming data-driven—said earlier this year that “good taste is more important than ever.” Nobody wants to be a data nerd anymore; we all want to be a tastemaker.⁴

Anecdotally, this seems especially true inside of AI companies, which are quickly becoming corporate trendsetters. The enthusiasm that SaaS startups had for analytics, experimentation, and rigorous quantitative thinking has been almost wholly replaced by a demand for people with taste and “agency.” One popular AI company—employing hundreds, used by millions, making hundreds of millions—has one person dedicated to data work. They have formal evals to measure how their product is performing, they said, but decisions are ultimately made based on how new features feel.

None of this is to say that data is going away. But it is falling out of fashion.⁵ It is fading into the background. In data we trusted; now, God is in the vibes.

Quantity has a quality all its own

But that story is incomplete. Even if vibe-driven decision-making is ascendant, how do we figure out those vibes? Business people need to know what’s happening with their businesses. Politicians need to know what voters think. Judges need to score a gymnastics routine. If not with numbers, then what?

On the second page of the Code of Points, there is an ad. It is for the Fujitsu 3D Sensing and AI Judging Support System:

The Judging Support System (JSS) employs a multi-step process to analyze gymnastic performances. It begins by capturing 3D data of the athletes’ movements without the use of physical markers. AI-powered pose estimation algorithms then identify and track the position of the gymnasts’ joints throughout their routines. …

To understand the details of gymnastic routines, the AI powering the JSS learned from a database of 8,000 routines. … The system can discern differences between elements, transitions, and pauses, as well as its understanding of the specific criteria for deductions based on deviations from the ideal execution. …

In addition to providing near real-time feedback to judges—identifying elements, calculating scores, and flagging deductions—JSS can be used by gymnasts and coaches to analyze performances and refine skills.

If AI is good at anything, it is good at interpreting the vibes. It is good at aggregating massive amounts of text—and increasingly, of video and audio—into its approximate average. Give it your support tickets and customer communications, and ask it questions about what it read. Don’t classify and categorize images; just ask an AI model what it thinks it sees. Don’t argue about steps on landings and timestamps when protests are filed; just have a robot watch the routine, compare it to thousands of others, and have it spit out how it deviated from the ideal execution. Ask it for the vibes.

The old generation might protest—this is not objective! This is not rigorous! There is nuance and bias in these questions, and “vibes” is just another word for hocus pocus punditry!

Maybe—though we would say that; it’s our whole bit. But regardless of our arguments’ merits, do we really think it’s going to keep winning? When a CEO asks us how a new product is doing, which answer will they prefer?

1. “Well, it depends on what you mean by ‘doing.’ Twenty percent of our existing customers have tried it, with a 7-day retention rate of 35 percent, and it varies by customer segment. New users have a 44 percent adoption rate, but only a 12 percent retention rate, though that’s probably skewed because we onboarded a large new customer last week. Moreover, we’re still investigating how adoption rates vary by signup source (with the usual caveats about attribution). Once the experiment is done in 494 30 days, we’ll have more details statistics we can share.”

2. “It’s not very good, because most customers say that it feels buggy.”

Is the second answer right? It almost doesn’t matter. As, uh, Stalin once said, “quantity has a quality all its own.”⁶ If you make something accessible and compelling enough—Wikipedia over Encarta; news on social media over news in the paper; ChatGPT over manual searches; immediate answers describing the vibes over legalistic answers about numerical minutiae—that’s where people will turn.

And then, it becomes self-reinforcing. The more that people use Wikipedia, the better it gets. Similarly, the more CEOs ask about their customers’ vibes, the more effort we’ll go to to understand them. When numbers were in demand, we built a giant, sprawling network of suppliers to collect, store, transform, and aggregate them. If the new bosses demand vibes, we’ll find the suppliers and build them a vibe stack.

Vibe native

Ask data people what teams need to do to prepare for the next era of analytics, and they will tell you about the importance of building semantic ontologies for AI agents. They will talk about context engineering, and metadata management, and the supreme importance of data quality. They will talk about the layers of technologies that need to get created to make sure that AI agents compute metrics consistently and accurately. They will talk about the various things that companies need to invest in tomorrow, so that data teams can make good on the promises that they made yesterday. All we need, they might say, is to finally build a better code of points.

It is so much work to do all of this, and it takes a lot of faith to believe in it. Why should the next generation will have that faith? They were raised on taste, aura farming, and ChatGPT’s instant answers. Their politicians—Trump, Mamdani—won on instinct and aesthetic. Their heroes on Twitter are the menswear guy, not Nate Silver. Their startups were more founder-mode than data-driven; their success came from their agency and not their analytical reasoning. They are vibe native.

For those of us who are still trying to build a data-driven world, that may be the only context that matters: We can keep trying to bring data, but in vibes they trust.

1. Their website has several menus, the first of which is called “General Information.” The first item on that menu is “Bank Details,” which contains, without explanation, all the information necessary to wire them money. Which, uh.

2. Four! One Mississippi two Mississippi three Mississippi four Mississippi! About that long!

3. Included in the lawsuit: Arguments about the Court of Arbitration for Sport sending important emails to the wrong address, very detailed discussions about video timestamps, and a debate about what it might mean that a video file was called “v3.”

4. As one commenter put it, “OpenAI might be winning with virality, but Anthropic could plant a flag squarely on the cultural edge and in a race this fast, taste can be power.”

5. Moneyball didn’t end scouting in professional sports, but it made scouts subordinate to the statisticians.

6. Or maybe it was a U.S. defense contractor?