The year in AI was crazy. So was this meeting of AI’s brightest minds.

They made small talk about everything from the hours at Elon Musk’s xAI (long) to the likelihood that foreign spies were inside top AI labs (high).

And one venture capitalist described in exhaustive detail the circular flow of money behind a recent investment by Nvidia into his AI startup, enrapturing the crowd with his explanation of how capitalism really works.

For years, the Conference on Neural Information Processing Systems—everyone just calls it NeurIPS—was attended by only the most dedicated scientists on the frontier of artificial intelligence. But what was once a small and proudly dorky academic gathering has become something more like AI-palooza.

This year’s conference was attended by more than 24,000 people, who invaded San Diego earlier this month and made it the epicenter of the tech world. After spending much of the year racing to remake civilization, they packed a giant convention center and marveled at displays flaunting the latest AI research between bites of Auntie Anne’s pretzels. At night, they filled restaurants, bars and taco joints with gossip about the industry that promises to transform the global economy.

The weeklong extravaganza was a sign of just how much this previously obscure and neglected corner of tech has changed since the start of the AI boom. Attendees who once gushed about ChatGPT whispered about OpenAI’s declaration of a “code red" as rival labs threaten its dominance. University researchers complained about how thoroughly the corporate world has taken over their academic field, only to ask one another how much money it would take for them to accept a job offer from a tech giant. And just about everyone was trying to figure out how long the boom times would last.

Their conversations were not limited to convention halls crowded with researchers wearing white conference lanyards. The real action took place during hundreds of happy hours, invite-only dinners and exclusive yacht parties where nerds have never been so cool.

One highlight of the social calendar was a party on the USS Midway sponsored by the top AI startup in Canada. Cohere, which sells AI software to businesses, had decked out the famous aircraft carrier with holiday lights, and NeurIPS attendees who climbed up several flights of stairs to reach the top of the ship were greeted with a massive dance floor.

But with this crowd, there wasn’t much dancing. As the music blasted, the young partygoers debated whether AI could fully replicate the human brain and if Sam Altman was a trustworthy person.

In private conversations, researchers whispered about which lab might be the first to build AI “superintelligence" that is smarter than mere humans—and whether they were in the middle of one humongous bubble.

They also groused about the growing ranks of venture capitalists, investment bankers, lawyers and consultants crawling the grounds, trying to cash in on the latest gold rush in Silicon Valley.

And some attendees were trying to lure those researchers away from Silicon Valley—to Wall Street. This year, the conference’s highest tier of sponsors included quantitative trading firms like Jane Street and Susquehanna that were searching for bright AI minds that might give them an edge in financial markets.

In many ways, it was a week that encapsulated the promise and peril of AI that defined the entire year.

It was a year when AI investments and startup valuations shot up—as did fears that the AI economy might come crashing down. In a single financial quarter, Microsoft, Alphabet, Amazon and Meta committed more than $100 billion to costly infrastructure. OpenAI, which loses money, made plans to raise lots of money and deals to spend even more. Nvidia, worth less than $500 billion when ChatGPT was released in 2022, cracked $5 trillion in market value before it dipped at the end of 2025.

As the year went on, Mark Zuckerberg began a recruiting blitz that roiled the industry, offering elite researchers unfathomably large sums of money to join his AI dream team inside Meta. For some, the pay could be worth billions. As rivals countered his poaching efforts with enticing packages of their own, the talent wars led to AI hires being paid more than NBA stars.

Perhaps the year’s most consequential development came on the eve of the NeurIPS conference, when Google released a version of its Gemini large language model that soared up the industry’s closely watched leaderboards. Soon afterward, the Slacks of OpenAI’s employees pinged with a message from Altman in which the chief executive declared a “code red" to improve ChatGPT.

For much of the following week, the company’s researchers were hundreds of miles away from San Francisco in San Diego.

Geeking out and getting rich

NeurIPS has long been a meeting of the most important people in tech who nobody outside tech had ever heard of—a place for researchers to share their latest findings with the few others who actually understand what they’re talking about.

When it was founded in 1987, its creators envisioned a wonky gathering of academics to swap ideas about how brains and machines learn. Their chosen field of AI research was the niche domain of eccentric professors and their graduate students. At that first meeting, only a few hundred scientists showed up.

As the conference ballooned over the years, NeurIPS became the site of some landmark moments in AI history. In 2015, attendees buzzed about the launch of a nonprofit known as OpenAI. In 2020, OpenAI won the Best Paper Award for its groundbreaking GPT-3 model. And in 2022, the company was once again the talk of the conference when it released its first viral product during NeurIPS. It was called ChatGPT.

The conference itself went through a similar transformation. It even rebranded its abbreviated name to “NeurIPS" from “NIPS" following complaints that the acronym contributed to a hostile environment for women in the field. At this year’s overwhelmingly male conference, there were rarely any lines for the women’s bathrooms.

With the official attendance tally reaching a new high, some longtime researchers bemoaned how unwieldy the conference had become. Not long ago, they could easily read the whole lineup of papers. Now they couldn’t see all 5,000 research posters even if they tried.

But the earnest academic inquiry that has long animated the conference could still be found inside the convention center.

Researchers shared their concerns about a world where machines could come to dominate humans and debated whether today’s methods of improving AI were enough to power tomorrow’s breakthroughs. They geeked out about continual learning, where AI gets smarter as it absorbs new information, and reinforcement learning, or “RL," a technique in which the models learn through trial and error.

There were dozens of workshops with titles impenetrable to people without Ph.D.s. At one weekend session, “Evolving RL-Based Reasoning Data in LLMs," a startup founder who specialized in that esoteric area of research asked a reporter why they were at such a boring talk.

Margot Wagner, a 29-year-old postdoctoral researcher at the University of California, San Diego, has been to NeurIPS four times. This time, she noticed a few changes.

For one, big companies weren’t as open about sharing truly novel research as they had been in the past. One explanation is that by the time that research is submitted, accepted and presented, it can be several months old, which the frontier labs now consider ancient history.

This year, there wasn’t a standout product or new research paradigm like there had been in previous years. And with more focus on the large language models driving the AI boom, Wagner observed there were fewer posters on neuroscience. “It’s interesting," she said, “because that’s the neur in NeurIPS."

A short walk away from the research papers was a sprawling expo hall lined with recruiting booths run by tech behemoths and the well-funded startups hoping to challenge them. At Google’s, there was a packed schedule of panels highlighting its latest research. Tesla showed off its Optimus robot, which wiggled its lifelike fingers to the crowd.

By night, the action moved to the luxury hotels where those companies held private receptions at rooftop bars and pools.

That was where the conversations got slightly less technical—and a lot more fun.

Ashwin Ramaswami, a 26-year-old Stanford graduate who co-founded an AI startup called Corridor, made a last-minute trip to San Diego after realizing that all of his friends from San Francisco would be there. His five days at NeurIPS left him feeling inspired by how much attendees still cared about discussing their work in the open, even as secretive labs have come to dominate AI research. “You can see where the roots of the AI revolution came from," he said.

He also made it to a house party with a wait list of more than 600 people, many of whom were trying to squeeze their way into a back-alley entrance, which left him with another takeaway.

“I never realized researchers partied so hard," Ramaswami said.

The hottest party one night was an invite-only cruise around the bay. Tickets were hard to come by and reserved for “those building, funding, and researching the technologies that define intelligence." As passengers sipped flutes of Champagne for hours on the triple-decker yacht, they played games of “Guess Who?" customized with the faces of AI legends.

Is your person Canadian? Did they win a Turing Award? Do they have curly hair?

Is it…Yoshua Bengio?

Over at another party on land, NeurIPS attendees were quietly asking another question: What’s your number?

They weren’t looking for each other’s contact information. The number these AI researchers had in mind was a dollar figure: the amount of money it would take for them to leave academia and join “industry," a word they uttered with a hint of disgust.

Even the academics who are not motivated by money find industry opportunities hard to resist. In addition to eye-popping salaries and stock options, they get access to the resources they really care about: advanced chips and computing power. As federal cuts threaten university funding, the increasingly lucrative offers from tech companies have only become more compelling.

When one researcher sheepishly named their number, it was $100 million.

A-list AI celebs

Inside the halls of the conference, AI celebrities who wouldn’t be recognized anywhere else in the world couldn’t walk around NeurIPS without being mobbed.

Jeff Dean, Google’s prominent chief scientist, gave multiple standing-room-only talks about the latest Gemini model. Among those standing at the back of the crowd taking notes were some researchers from OpenAI. The day before, attendees in backpacks huddled around Richard Sutton, a pioneer of reinforcement learning who was also carrying a backpack. And perhaps nobody at the conference was stopped for more selfies than podcast host Lex Fridman, who has interviewed everyone in AI from Altman to Zuckerberg.

“It wasn’t just selfies," he said. “There was a lot of kindness, love and super-technical research conversations."

As the programming came to an end, packs of AI researchers began heading back to the convention center for the closing ceremony. But when a freight train stalled at a key intersection in the middle of the city, they found themselves stuck on the wrong side of the street for nearly 30 minutes. They didn’t know what to do—or where to go. If they tried slipping through the opening between cars, the train could start moving again at any moment.

It was as if they were staring at the AI industry’s broader dilemma: whether to move boldly into uncertain territory or wait for clarity and risk falling behind.

Some impatient researchers climbed through the train cars and crossed the street without incident. Others decided to walk along the mile-long sidewalk to the front of the train.

Those who chose the dimly lit path happened upon their version of an A-list celebrity: the man known as the godfather of AI.

Geoffrey Hinton has been a presence at NeurIPS since its very first meeting, when the future Nobel Prize winner was publishing work that laid the foundation for the modern era of AI. He’s attended the conference many times over the years, presenting seminal papers and picking up awards for his research. After quitting his job at Google two years ago, he has become an outspoken critic of AI’s growing risks and even expressed some regret for his life’s work.

Hinton was surrounded by conference attendees hanging on his every word as they walked down that path into the darkness ahead.

Write to Berber Jin at berber.jin@wsj.com, Meghan Bobrowsky at meghan.bobrowsky@wsj.com and Ben Cohen at ben.cohen@wsj.com