Author: openjargon

• Yann LeCun, Meta's chief AI scientist, is leaving the company.
• LeCun said in a social media post that he is creating a new AI startup.
• The departure comes amid a period of instability within Meta's AI organization.

Yann LeCun, Meta's chief AI scientist and one of the most influential figures in the field, is leaving the company to start a new AI venture, a Meta spokesperson confirmed to Business Insider.

LeCun announced the move on Facebook, sharing that he is building a startup centered on his long-standing interest in world-model research. Meta will partner with LeCun on his new venture, a Meta spokesperson told Business Insider, but didn't reveal any details about the nature of the partnership.

LeCun's departure comes during a period of instability inside Meta's AI organization. Over the past few months, Meta has hired dozens of top researchers and engineers from rivals and reorganized its AI efforts under the new Superintelligence Labs division, led by former Scale AI CEO Alexandr Wang. Tensions emerged within the newly formed team between the highly compensated new hires and the existing researchers, some of whom have threatened to quit, Business Insider previously reported.

In August, the company made the biggest reorganization of its artificial intelligence operations to date, creating four distinct teams that focus on research, training, products, and infrastructure.

This shift followed the company's pivot toward out-competing OpenAI, Google DeepMind, and Anthropic on large-scale AI models. At the same time, researchers have cycled in and out of key roles, and Meta's Llama 4 release drew muted reactions internally and externally. Earlier this week, Souminth Chintala, the creator of Meta's open source AI framework PyTorch, left the company after 11 years to join Mira Murati's Thinking Machines Lab.

LeCun's departure isn't entirely surprising. He's been a vocal critic of relying too heavily on large language models, arguing instead for his JEPA approach — a method that trains AI to understand and predict the physical world from images and sensory data, rather than generating text. Meta, meanwhile, has increasingly focused on scaling LLMs and pushing commercially driven model development.

Have a tip? Contact Pranav Dixit via email at pranavdixit@protonmail.com or Signal at 1-408-905-9124. Use a personal email address, a nonwork WiFi network, and a nonwork device; here's our guide to sharing information securely.

• Google's Gemini 3 AI impresses with advanced multimodal and coding capabilities.
• Gemini 3 enables users to create interactive websites, simulations, and video games easily.
• Strong reviews for Gemini 3 have boosted Google's market value close to Microsoft's.

Business Insider's amazing Google reporter Hugh Langley has been playing (er… working) with Google's new Gemini 3 AI service this week.

This is the latest big AI model release, competing with OpenAI's GPT-5, xAI's Grok 4, and the latest offerings from Anthropic.

Gemini 3 is getting good reviews so far. So good that Google shares hit a record on Wednesday, putting the company's market value very close to Microsoft.

Here are Hugh's initial thoughts, after trying Gemini 3 out for a day or so:

"I think where Gemini 3 is most impressive—and where it's already grabbing attention—is its ability to create new things from whole cloth, thanks to improved multimodal and coding capabilities."

Gemini 3 is particularly good at designing interactive simulations. That could be interesting for visual learning, building websites and apps, or just having fun, Hugh told me.

He played with this new model in a Google sandbox called AI Studio. It's like a real sandbox, but for developers. And it's digital, giving access to Google's AI offerings via the internet. There's no real sand.

Hugh started with something basic: an interactive website about elephants.

"I asked Gemini to include lots of fun widgets and trivia about the animal, but little else. I wanted to see how much Gemini would fill in the gaps."

While the overall website design was a little sparse, it delivered. Hugh liked this little widget that generated a fun elephant fact every time he pressed a button.

"It also included a mini game where I had to feed the elephant by giving it peanuts," Hugh said. "Once I filled the bar, a pop-up message informed me the elephant was now happy, so that's nice."

Being able to visualize complex ideas is an area where AI could be particularly useful. Hugh asked Gemini 3 to create an interactive website to explain photosynthesis.

"Gemini generated some sliders to adjust the levels of sunlight, water and carbon dioxide that were floating around as different colored particles," he said. "If I got the balance correct, it told me I had successfully created energy!"

Hugh hadn't generated energy. In fact, that Gemini 3 model run probably sucked up quite a bit of power. Still, pretty impressive.

Other Gemini 3 users are creating interesting new things. Hugh liked this idea to have Gemini build a Lego creator, through a simple prompt such as "Create a 3D Lego builder. Let me select different shapes of brick."

Here's an example. Hugh was able to replicate his own version.

Then, there are all the video games that users are getting Gemini 3 to create. Jeff Dean, one of Google's top AI researchers, posted several clips of these games.

Hugh gave this a shot, too. This required a bit more back-and-forth with Gemini to get right.

He wanted to make "Super Dario Land," a game where the player has to get Anthropic CEO Dario Amodei into the correct warp pipe. If they succeed, the player is rewarded with AGI (the theoretical moment when machines outperform humans on most tasks).

Hugh asked Gemini to style it like one of Nintendo's old Game Boy games.

"At first, Dario couldn't jump high enough, so I asked Gemini to fix that," Hugh told me. "With one extra prompt, the physics were solved!"

The game was very quickly playable as Gemini did the work of mapping the controls to Hugh's computer keyboard without any direction from him.

"The game itself might not be a hit, but I'm leaving the door open for a sequel. Dario, call me!" Hugh said.

If anyone wants to play Hugh's game, send him an email at hlangley@businessinsider.com. It's fun!

Sign up for BI's Tech Memo newsletter here. Reach out to me via email at abarr@businessinsider.com.

• Elon Musk predicted a future where money will "stop being relevant," thanks to AI and robotics.
• Speaking at the U.S.-Saudi Investment Forum, Musk also said that work would become "optional."
• Musk has previously said that AI robotics, like Tesla's Optimus, would eliminate poverty.

In Elon Musk's future, we won't need jobs or money, and there will be no poverty.

At the U.S.-Saudi Investment Forum on Monday, where Musk sat on a panel with Nvidia CEO Jensen Huang, Musk said that money would "stop being relevant" thanks to AI.

"There will still be constraints on power like electricity and mass," Musk said. "But I think at some point currency becomes irrelevant."

He linked it to the books of science fiction author Iain Banks, who wrote the Culture series between 1987 and 2012. Those books help "get a sense for what a probable positive AI future is like," he said.

Musk also mentioned the end of work itself, saying that it will be "optional," like "playing sports or a video game."

He compared the future of work to gardening. "It's much harder to grow vegetables in your backyard, but some people still do it because they like growing vegetables," he said."That will be what work is like: optional."

Over the past few months, Musk has shared his vision for a future with AI. That includes ending poverty, something he described at a recent investor meeting.

"People often talk about eliminating poverty, giving everyone amazing medical care," Musk said at the shareholder event earlier this month. "There's actually only one way to do that, and that's with the Optimus robot."

When AI and robotics, like Tesla's Optimus, eliminate all work and money, the government should hand out a universal income, Musk told Joe Rogan in October.

That income shouldn't just be a universal basic income — it should be a universal high income, he said.

"We'll have, in a benign scenario, universal high income," Musk said. "Anyone can have any products or services that they want. But there will be a lot of trauma and disruption along the way."

After describing the future irrelevance of money at the U.S.-Saudi Investment Forum, Musk ribbed his panelmate, Nvidia's Huang.

"By the way, the Nvidia earnings call is today," Musk said. Nvidia will report its third-quarter earnings after the closing bell.

"Since currency is irrelevant…," Huang said.

"Cheers," Musk said. The CEOs clinked their bottles of Acqua Panna.

• US Coast Guard cutter Stone offloaded a substantial amount of cocaine seized from drug interdictions in the Pacific.
• The deployment included 15 interdictions, three of which occurred within a single night.
• A new drone capability helped the Stone track drug smuggling vessels.

PORT EVERGLADES, Florida — A US Coast Guard crew successfully seized a record-breaking amount of cocaine from drug runners with the help of an unusual reconnaissance drone.

The unique tail-sitter drone, capable of taking off and landing vertically, allowed the crew of the Coast Guard cutter Stone to easily put eyes in the sky and spot drug smugglers during a recent deployment in the eastern Pacific.

The crew of the Stone, a large Legends-class National Security cutter, offloaded over 49,000 pounds of cocaine worth more than $362 million at Port Everglades in Fort Lauderdale, Florida, on Wednesday. It was a landmark offload marking the most cocaine seized by a single Coast Guard cutter during a deployment. The majority of the cocaine taken during this deployment came from Colombia, officials said.

"What you see behind me is more than just a pile of cocaine," Vice Adm. Moore, commander of Coast Guard Atlantic Area, said, standing amid cocaine packages stacked in long rows on the Stone's deck. "It represents a tangible victory in our ongoing fight against transnational criminal organizations and narcoterrorism."

The deployment, which began in August, was part of the Coast Guard-led Operation Pacific Viper targeting drug-running operations in the eastern Pacific. Coast Guard officials say the service is accelerating its counter-narcotics missions, resulting in record numbers of drug interdictions. In fiscal year 2025 alone, the Coast Guard seized almost 510,000 pounds of cocaine, the most in its history.

The Stone's recent deployment in the Pacific included 15 interdictions, three of which occurred on the same night.

The three vessels were spotted in rapid succession by a new capability on the Stone, Shield AI's MQ-35 V-BAT. The uncrewed aerial vehicle, which was operated by a contractor team, spotted the first boat in dark waters during the night, prompting the Stone to prepare a boarding team.

Capt. Anne O'Connell, the commanding officer of the Stone, told Business Insider that as the team and the armed interdiction helicopter were interdicting the vessel, the V-BAT went out to patrol the area further. "That's when they saw the wake from the second TOI," or target of interest, she said.

As the second boarding team went out, the drone set out on another patrol, finding a third boat nearby. That night, a total of 12,000 pounds of cocaine were seized, along with seven suspected narcotics traffickers.

The drone, O'Connell said, was integral to the operations that night because it allowed the Stone's crew to continue monitoring surrounding areas while completing boarding processes, which can take anywhere from two to eight hours depending on the size of the vessel and the complex law enforcement procedures that Coast Guard teams must follow.

The V-BAT flies over a designated area determined by the Coast Guard. The drone's operators receive specific instructions on what to look for, and once it's airborne, its live video feed is transmitted to the ship, where crew members can watch it on monitors.

This was the Stone's first deployment with the V-BAT, and it's also one of the first cutters to have it on board, O'Connell said. Its usefulness was especially notable in the large operating area of the eastern Pacific, as the uncrewed aerial system could make up for a lack of fixed-wing aircraft doing reconnaissance.

The V-BAT is an unusual drone design, featuring ducted-fan technology for lift. Built to have a small tactical footprint, according to its maker, the drone can take off in winds up to 25 knots from vessels on the move at up to 10 knots. Shield AI says it can offer over 13 hours of flight time for persistent surveillance.

The company notes that a two-man team can have the V-BAT assembled and operational in under 30 minutes.

While the V-BAT, like other capabilities making the Stone a premier vessel for these types of missions, proved valuable, officials said credit belongs primarily to the crew.

"All of those elements, with the UAS and our HITRON [Helicopter Interdiction Tactical Squadron] and our small boats," O'Connell said, are incredible capabilities. "But the secret sauce is our people, and they are what makes us successful."

• My partner is 58, and I am 69, and our kids have all left our home.
• We are between parenting and becoming grandparents, so we decided to explore the world.
• We call it rotational living, and are giving ourselves time to figure out where to live next.

My partner Deb, 58, and I, 69, are taking a gap year that began in January 2025. Except, we have no hard ending — we call it rotational living.

We gave up our fixed address, a rental too large and expensive for our needs, while we explore the world. We spent several weeks in São Paulo, about the same amount of time in Providence, and a few days in rural Vermont, watching spring settle into the mountains. We lived in Montreal for a month and a half this summer and depart for three months in Brazil again the first week in December. We look forward to Europe, South Africa, and the Middle East next year.

Rotational living works for us in part because we both prefer travel as though living in a new place. When one trip ends, another begins, in a different city, state, or country.

We learned a lot from our travels

We've learned a few things while traveling together. First, you have to truly enjoy your partner. We're both writers and consultants, so we create work in similar ways. We've realized how much we rely on the predictable — the same coffee maker, clocks you don't have to search for, all the different shoes — and how exhausting unpredictability can become.

We accommodate these challenges by going to bed early and leaving room in our days for uncertainty. These few negatives of rotational living don't detract much from the pleasures. Not knowing lies at the core of exploration, and so we have learned to master uncertainty, embrace adventure, and love freedom.

We had talked about living this way for years. Then my mother died in the fall of 2024 after a long and glorious life. The kids — my two in their 30s, Deb's three in their 20s — have launched, all of them in careers they trained for, and none of them have children yet. This little gap, between parenting and grandparenting, arrives like a gift. We look forward to becoming our future grandkids' default babysitters and embracing a fixed address when that moment arrives. But in the meantime, we contemplate where to go next.

We are spending less money

Our decision also has a financial underpinning, although the professional freedom Deb and I enjoy might have led us to this choice anyway.

We lived outside Boston, in a community for people who moved there for the top-notch schools. Our large apartment cost an absurd sum compared to our needs, but nothing within a two-hour radius saved us much money. In truth, we don't yet know where we want to live, so rather than spend thousands a year on rent for a place we don't love, why not spend less and live everywhere?

Calculating the cost of rotational living clarified that a conventional home led to a life at the edge, whereas rotational living brought us the luxuries we most desire: learning new cultures, eating well, time with friends and family, and artistic inspiration. We started a blog called Breakfast: A Love Story to share this joy with the world.

When we visit Brazil, we can rent comfortable apartments for under $1,000 a month, pretty much whenever we want to go. The same goes for India, another destination on our list. We're looking for an open month for Europe, where we will mostly stay with friends. Work obligations sometimes set our travel map. Just as often, we imagine places we want to experience, such as Japan and Australia. The moment we make friends in those places, we will go.

Andy Hoffman began writing professionally as a teenager and has founded several businesses, largely in educational technology. He currently lives everywhere.

• OpenAI revenue growth is surpassing its own forecasts, according to a deep dive by Barclays.
• New revenue streams like advertising and AI agents could boost OpenAI's revenue.
• These signs of acceleration could fuel AI infrastructure spending, the analyst said.

OpenAI is "running ahead" of its own revenue targets, a signal that the company driving the generative AI boom is expanding faster than even its backers expected.

That's according to a new deep dive from Barclays tech analysts, led by Ross Sandler. They wrote this week that OpenAI's better-than-expected growth trajectory reinforces the AI infrastructure investment wave rather than slowing it, despite mounting concerns over capital intensity and potential market bubbles.

OpenAI's revenue performance is roughly 15% above 2025 forecasts and 50% ahead of 2027 projections, according to analysts' estimates, based on CEO Sam Altman's recent comments that the company is on pace to reach $100 billion in annual recurring revenue by 2027. That's about a year earlier than previously expected.

The Barclays analysts attributed the outperformance to user growth, steady conversion from free to paid subscribers, and the rapid scaling of OpenAI's enterprise and application programming interface (API) businesses.

Their research note outlined key performance indicators that OpenAI must hit to keep this revenue momentum going:

• Maintaining a 50 million monthly increase in weekly active users (WAUs)
• Keeping free-to-paid conversion rates near 4%
• Growing average monthly revenue per user from $30 to $55 through new, higher-tier offerings
• The API business, which provides access to GPT models, needs 6x growth
• New sources of revenue must emerge, such as advertising and AI agent services

If ChatGPT grows to about 2 billion weekly active users by 2028, that could help OpenAI generate $100 billion in annual recurring revenue, depending on how many of these users subscribe to paid versions of the chatbot service, the analysts estimated.

The research note also pointed to new revenue streams, including advertising on the ChatGPT free tier and an emerging "Agents-as-a-Service" model (effectively digital employees that can handle tasks for businesses). The analysts say both businesses could meaningfully expand monetization over the next two years, while the API business continues to grow as adoption broadens.

There's also a shopping referral fee revenue stream that comes with OpenAI's recently launched Instant Checkout feature in ChatGPT, the analysts wrote.

This revenue expansion means increased compute demand. OpenAI's compute budget is now projected to exceed $450 billion from 2024 through 2030, with total obligations of around $650 billion, some of which extend beyond 2030, according to Barclays research.

The analysts wrote that these signs of acceleration, rather than signaling a coming slowdown, could extend the AI investment supercycle.

"We would expect the other labs to continue to keep their foot on the gas," Sandler and his colleagues wrote in their note this week. "And hyperscalers are likely to keep their spending levels up, despite concerns."

Sign up for BI's Tech Memo newsletter here. Reach out to me via email at abarr@businessinsider.com.