New post

RESEARCH

Every major AI model can ace a creativity test on its own. They also all ace it the same way.

Researchers at Duke University and the Technion–Israel Institute of Technology tested 22 large language models — from OpenAI, Google, Meta, and Mistral — on three standardized creativity tests alongside 102 human participants. Each model individually scored as well as or better than the average human. But when the researchers compared AI outputs to each other, the picture changed: AI responses were far more similar to one another than human responses were, across every test.

The gap wasn't subtle. On one measure, the effect size hit 2.2 — nearly three times what social scientists consider "large." The reason is straightforward: every major model trains on essentially the same data and optimizes for the same objective. The result is output that looks creative on the surface but is deeply homogenized underneath. And switching models doesn't help — going from Google to Meta produced barely more variety than switching between two models from the same company.

This is already showing up in the real world:

+ In fashion, 34% of executives say their teams use generative AI for marketing copy, accelerating sameness at scale.

+ In travel, AI-built booking platforms keep popping up with near-identical language and design, making it harder for real operators to stand apart.

As Duke's Emily Wenger put it: "If you're trying to come up with an original concept or product to stand out from the crowd, this work highly suggests you should bring together a diverse group of people to brainstorm rather than relying on AI."

Companies that lean hardest on AI for creative work are quietly erasing the thing that made them different. Every major model trains on the same internet and optimizes for the same goals, so the more teams rely on them without adding something of their own, the more everything blends together. AI was supposed to democratize creativity. It might just be standardizing it.

WORKFORCE

Gen Z uses AI just as much as last year. They like it a lot less.

A Gallup survey found that 51% of Gen Zers use generative AI at least weekly — basically unchanged from a year ago. But excitement about AI dropped 14 points to just 22%, while anger surged 9 points to 31%. Even among daily users, excitement fell 18 points year over year. That breaks the usual pattern for tech adoption, where heavier use breeds more enthusiasm. Gen Z is doing the opposite.

The oldest Zoomers are the most skeptical because they're old enough to watch AI reshape the job market they haven't fully entered yet. Nearly half of employed Gen Zers now say AI's risks outweigh its benefits, up from 37% a year ago.

But their concerns go deeper than losing a job. Gen Z is making pointed judgments about what AI does to the way they think:

+ Critical thinking: 42% say AI is harmful to their ability to think carefully, versus 25% who call it helpful.

+ Creativity: 38% say AI hurts their ability to come up with new ideas on their own.

+ Learning: 80% believe AI tools will make it harder for them to learn in the future.

That skepticism is already showing up at work. Two years ago, employee complaints were mostly about displacement — "Why am I training the robot that's going to take my job?" That specific fear has faded, but the resistance hasn't. Employees are quietly ignoring AI tools or working around them because they don't buy the strategy behind them. And it doesn't help when leaders like Anthropic's Dario Amodei warn AI could eliminate half of entry-level white-collar jobs. For a generation watching the ground shift, those statements feel more like threats than forecasts.

The tech industry assumed Gen Z would adopt AI like they adopted everything else: fast and with minimal friction. What the Gallup data shows is a generation fluent enough in technology to see exactly what AI threatens, and they're watching it reshape a job market they haven't fully entered yet. Companies building their AI strategies around the idea that younger workers will naturally lead adoption might want to check whether that assumption still holds. The workers who grew up on the internet aren't afraid of new technology. They're just not convinced this particular technology is on their side.

HARDWARE

Last weekend at the Beijing Half Marathon, Honor's "Lightning" robot finished the 21-kilometer course in 50 minutes and 26 seconds — nearly seven minutes faster than the human world record. No remote control. No human operator. Fully autonomous.

It was one of roughly 300 humanoid robots from 26 brands that lined up alongside human runners in what's quietly become the robotics industry's most revealing annual showcase. A year ago, the winning robot time was 2 hours and 40 minutes. This year: under 51 minutes. That's nearly two-thirds faster in 12 months.

It wasn't exactly graceful — one robot fell at the starting line, and about 60% of the 102 teams still relied on remote control rather than full autonomy. But the spectacle is overshadowing a more significant shift. The money pouring into humanoid robotics has started to look less like VC enthusiasm and more like early-stage industrialization:

+ Apptronik closed a $520 million extension, bringing its total Series A to $935 million — one of the largest early-stage robotics rounds ever.

+ Figure raised over $1 billion in its Series C at a $39 billion valuation.

+ AGIBOT, a Chinese robotics company, was ranked No. 1 worldwide in humanoid robot shipments for 2025 by research firm Omdia.

China is pushing especially hard. The government published a five-year blueprint to fast-track humanoid robot adoption in manufacturing, and dozens of Chinese firms are racing to move from lab demos to commercial deployments. But most of the industry is still in the earliest stages — somewhere between prototype and pilot.

The gap between finishing a half marathon and doing useful work remains enormous:

+ Battery life still tops out at two to three hours per charge.

+ Cost needs to drop to roughly $42,000 before enterprises see a reasonable ROI — cutting-edge research models still run into the millions.

+ Dexterity is still a challenge — tasks that feel instinctive to humans, like gripping an oddly shaped object, can still trip up machines.

Wang Xingxing, CEO of Unitree Robotics, has a clear benchmark for when the industry has truly arrived: "A true 'ChatGPT moment' for the sector will come when robots, placed in unfamiliar real-world settings, can complete about 80 percent of tasks simply by following voice or text instructions."

A year ago, the fastest robot in Beijing needed nearly three hours to finish a half marathon. Now one can beat the human world record. That kind of improvement doesn't happen in a field that's stalling out, and the billions flowing in suggest investors agree. The question is whether the factory floor catches up to the racetrack before the hype cycle turns, because right now, the robots are running faster than the business model.

IN OTHER NEWS

+ OpenAI to spend $20 billion on Cerebras chips and take an equity stake in the startup — a direct shot at Nvidia's dominance

+ Google is reportedly in talks with Marvell to build two new custom AI chips — one a new TPU, the other a memory processor to supercharge it

+ TSMC breaks its own rules, ramping up 3nm chip production simultaneously in Taiwan, the U.S., and Japan to meet explosive AI demand

+ Cerebras files for a public offering, positioning itself as a real challenger to Nvidia after landing the massive OpenAI deal

+ Cloudflare launches a major expansion of its Agent Cloud platform for autonomous AI agents — including sandboxes where agents can build and deploy code on their own

+ Microsoft and Stellantis sign a five-year AI partnership covering 100+ AI initiatives across Jeep, Peugeot, and more

+ Global memory chip shortages peaked in Q1 2026, stalling AI infrastructure rollouts as agentic workloads devour DRAM faster than anyone expected

WHO'S HIRING IN AI

+ Airbnb — Senior Manager, Product & AI Policy

+ Netflix — Creative Evaluation Lead, Artwork (AI x Promotional Art)

+ Mercury — Senior Product Manager, API & Agentic Banking

+ Abbott — Senior Director, AI Strategy, Governance and Transformation

AI OR REAL?

Which image is real?

Option A | Option B

AI TOOLS

+ Littlebird (sponsored): Think of it as an AI that actually knows what you're working on. It watches your screen, takes notes in your meetings, and remembers all of it. So when you forget where you saw something, you just ask.

+ Clico (sponsored): A free add-on that puts a writing helper directly inside Gmail, Google Docs, LinkedIn, and wherever else you type — no more copying your email into another tab and pasting the answer back.

+ Gemini: Google's AI can now pull from your Google Photos library to generate personalized images — describe a scene with your family and it figures out who's who from your labels

+ Gemini Notebooks: Google just made its persistent AI workspace free for everyone — save conversations, files, and context in one place so Gemini remembers your projects across sessions

+ Gemini Live: Google is redesigning its real-time voice AI on Android, ditching the fullscreen takeover for a sleeker in-app interface with camera sharing and screen sharing built in

+ Google AI Mode in Chrome: Chrome now opens websites side-by-side with AI answers instead of replacing them, and can search across all your open tabs at once