Everyone’s excited about the wrong AI breakthrough
Plus: AI agents caught faking their work, Neom’s spectacular collapse, and why we’re in AI’s “dial-up era”
Dear Friend,
Two very different visions of AI’s future emerged this week, and they tell us everything about where we’re headed: OpenAI launched ChatGPT 5.1, complete with “personalities” you can toggle like ringtones — professional, cynical, cheerful, you name it. Think of it as AI cosplay: same engine, different masks.
Meanwhile, AI pioneers Fei-Fei Li and Yann LeCun, independent from each other, announced they’re tackling something far more fundamental: teaching AI to actually understand the world. Both will focus their energy on building “world models” that give AI a mental map of how reality works – the same way a toddler learns that objects don’t disappear when you hide them behind your back.
It’s quite the contrast - one approach asks: “How can we make AI sound different?” The other asks: “How can we make AI think different?”
My money is on Fei-Fei and Yann…
P.S. You may have seen it in the email on Wednesday – we launched the radical Pulse and would love to hear from you. What’s keeping you up at night, what’s energizing you, where you’re headed – it takes 2-3 minutes to respond, and we will share the results back with you. Take the pulse now.
P.P.S. If you want to hear something that’ll make you rethink how big companies actually work, check out the latest Disrupt Disruption episode: “We Couldn’t Rebuild Our Own Success Today: HP Fellow Will Allen on Why Big Companies Kill Innovation, Research vs. Development, and Bringing AI to the True Edge.” Will’s perspective on why scale becomes a cage—not a moat—stuck with me. I think it’ll stick with you too.
And now, this…
Headlines from the Future
Agents Fabricate Data to Hide Their Failures. Researchers at CMU compared 48 human workers against four AI agent frameworks across sixteen realistic work tasks—data analysis, engineering, writing, design. The agents were 88% faster and cost 90-96% less than humans. Sounds great, until you look at how they work: agents fabricate plausible data when they can’t complete tasks, misuse advanced tools to mask limitations (like searching the web when they can’t read user-provided files), and take an overwhelmingly programmatic approach to everything – even visual design tasks where humans use UI tools. In essence, we’re training agents optimized for appearing productive rather than being accurate – and they’re learning to fake competence at 90% lower cost. Bravo!
↗ How Do AI Agents Do Human Work?
━━━━━
Neom Isn’t Hot Anymore. While Saudi Arabia’s desert is heating up, Neom – the kingdom’s grandiose plan to create a city fit to house the future of humanity – seems to crumble.
A new report from the Financial Times cites high-level sources within the project to paint a picture of dysfunction and failure at the heart of the quixotic effort.
No one is surprised. This hits particularly hard:
One such addition is an upside-down building, dubbed “the chandelier,” that is supposed to hang over a “gateway” marina to the city: As architects worked through the plans, the chandelier began to seem implausible. One recalled warning Tarek Qaddumi, The Line’s executive director, of the difficulty of suspending a 30-storey building upside down from a bridge hundreds of metres in the air. “You do realise the earth is spinning? And that tall towers sway?” he said. The chandelier, the architect explained, could “start to move like a pendulum”, then “pick up speed”, and eventually “break off”, crashing into the marina below.
↗ Saudi Arabia’s Dystopian Futuristic City Project Is Crashing and Burning
━━━━━
Fei-Fei Li’s Bet. Legendary AI researcher Fei-Fei Li just published her perspective of where AI is/needs to head next: Spatial Intelligence. While today’s large language models (LLMs) are “wordsmiths in the dark”, eloquent but without real-world grounding, the future of AI lies in understanding and interacting with the physical world just as we do.
Spatial intelligence will transform how we create and interact with real and virtual worlds—revolutionizing storytelling, creativity, robotics, scientific discovery, and beyond. This is AI’s next frontier. […] Spatial Intelligence is the scaffolding upon which our cognition is built.
↗ From Words to Worlds: Spatial Intelligence is AI’s Next Frontier
━━━━━
AI’s Dial-Up Era. Here is a well-reasoned and insightful article that uses a useful historical analogy to frame the current moment in AI. The central thesis – that we are in the “dial-up era” of AI and that its long-term impact is being debated with the same shortsightedness as the early internet – is a compelling one. Read the whole thing.
━━━━━
The Robotaxi Is (Finally) Here To Pick You Up. Waymo, the self-driving car division spun off from Google [X], just announced that its cars are now allowed to drive on the freeway in places like San Francisco – which is a huge step forward for the company, as it allows Waymo to operate cars picking you up at SFO airport and drive you right into downtown SF. Here is a good explainer why this is such a big deal and why it took so long:
↗ Waymo is finally ready for freeway service
What We Are Reading
🤦 New Research Warns That AI Is Causing a ‘Reverse Dunning-Kruger Effect’ AI experts are the worst at spotting their own mistakes, blindly trusting chatbots despite knowing better. @Jane
🚗 Waymo Robotaxis Are Now Giving Rides on Freeways in LA, San Francisco, and PhoenixEngineers started working on what is now Waymo 16 years ago. Fast forward and the robotaxis are now taking on highways. @Mafe
⚖️ The Progress Paradox A politics of technology and entrepreneurship that favors private monopolies might support innovation, while thwarting broader measures of social progress. @Jeffrey
🌫️ Jargon Is Hurting Your Strategy When your business architecture rests on vague jargon, the language that should build clarity instead undermines your strategy before it even lands. @Kacee
🏢 Here’s How the AI Crash Happens It’s not that the hyperscalers are spending money at an alarming rate that could collapse the AI bubble; it’s the data centers. It always comes down to real estate. @Pascal
Down the Rabbit Hole
🌌 Universe’s expansion ‘is now slowing, not speeding up’
🚢 China keeps winning: China unveils thorium-powered nuclear cargo ship that can carry 14,000 containers
🧸 The ‘Toy Story’ You Remember
🤖 Investors’ ‘dumb transhumanist ideas’ setting back neurotech progress, say experts
🚸 Another country joins the social media ban for kids and teens: Denmark to ban social media for children under 15
☕ As a coffee lover, my heart just jumped a beat: Coffee has ‘astounding’ impact on irregular heartbeat in world-first trial
📷 Curious how the different AI models compare when it comes to image generation? We ran over 600 image generations to compare AI models
⌚ The history of Casio’s digital watches
💡 Using Street Lamps as EV Chargers
👨🏼💻 Laptop stickers are a thing. Here is a gallery
🌐 On the heels of Fei-Fei Li’s announcement (see above), this: Meta’s Chief AI Scientist Yann LeCun To Depart And Launch AI Start-Up Focused On ‘World Models’
🪙 End of an era: The last-ever penny was minted today in Philadelphia
Pascal’s geeky side lit up when he saw Valve’s announcement for their next generation SteamMachine.