!
👋 Dear Dancing Queens and Super Troupers,
This week, AI stopped pretending. No more fairy-tale storytelling about a “magical assistant that wants the best for you.” The masks slip, the cables show, and the real nerves are finally visible. Apple, OpenAI, Google, Xiaomi… each moved a piece, and the chessboard reshaped itself right in front of us.
Let’s start with the symbolic shock. Apple, the company that loves to explain it does everything “differently,” has chosen Google to serve as the cognitive foundation of its future models. Gemini becomes the engine of the next version of Siri.
This isn’t a cosmetic collaboration. It’s a strategic admission. Apple keeps the interface, the narrative, the privacy promise. Google supplies the brain. In Silicon Valley, they call that “focusing on core strengths.” In real life, it’s acknowledging who won the model war.
Meanwhile, OpenAI is advancing on two fronts. On one side, the futuristic vision with its “Sweetpea” AI earbuds, worn behind the ear, always on, always listening to the world. An assistant that no longer lives inside an app, but inside your environment.
On the other, a far more down-to-earth decision: ads are coming to ChatGPT. Not inside the answers, we’re told, but just underneath. Like YouTube in its early days. Like Google yesterday. Like every platform that’s realized inference is expensive and idealism doesn’t pay GPU bills.
And while some are discovering that AI needs cash flow to survive, others are changing scale, literally. In Germany, researchers are running a simulation on a supercomputer equivalent to the human cortex, with tens of billions of neurons and hundreds of thousands of billions of connections.
This isn’t “creating a mind” or “copying the brain,” but it is a new microscope. A tool to observe what only appears at large scale, where small models stop being representative and unexpected behaviors emerge.
Finally, Xiaomi plays the most pragmatic card of all. Audio glasses with no camera, no screen, no dystopian promises. Just sound, comfort, decent battery life, and a price that doesn’t make you sweat.
Here’s this week’s lineup :
👉 Apple hands Siri’s brain to Google 🤯
👉 OpenAI bets on wearables for the post-smartphone era ⌚
👉 Free AI is ending: OpenAI moves to ads 💸
👉 Simulating 20 billion neurons finally becomes possible 🧠
👉 Xiaomi proves AI can also stay invisible 👀
If we've forwarded this letter to you, subscribe by clicking on this link !
⚡ If you have 1 minute
Apple has decided: future Apple Foundation Models will rely on Gemini, Google’s AI. Siri will become more personalized as early as this year, while remaining embedded in Apple’s ecosystem and privacy promise. The implicit message is crystal clear: Apple keeps the product, Google provides the brain.
OpenAI is reportedly working on Sweetpea, a behind-the-ear wearable designed to compete with AirPods. Larger battery, dedicated chip, environmental sensors: the assistant no longer lives in an app, but in the real world. A release is mentioned around 2026, with very high industrial ambitions.
OpenAI will test ads in ChatGPT for free accounts and the Go subscription. Ads will appear below answers, never on sensitive topics, and will remain theoretically separate from “organic” responses. To avoid ads, you’ll have to pay more.
In Germany, researchers are using the JUPITER supercomputer to simulate a network equivalent to the human cortex: 20 billion neurons, 100 trillion connections. This isn’t artificial consciousness, but a new scientific microscope. At this scale, behaviors emerge that small models never reveal.
Xiaomi is selling Bluetooth audio glasses in Europe for €179. No camera, no screen, just sound, 13 hours of battery life, and a featherweight design. A utilitarian, reassuring approach: AI becomes discreet, portable, and socially acceptable.
🔥 If you have 15 minutes
1️⃣ Apple Hands Siri’s Brain to Google
The summary : Apple is now entrusting Siri’s intelligence to Gemini, Google’s AI model. This multi-year partnership aims to accelerate Siri’s overhaul, which had been pushed back to early 2026 due to insufficient internal progress. Gemini becomes the technological foundation, while Apple retains control over the user experience, deployment, and part of the processing via its private cloud.
Details :
A clear-eyed technological admission: Apple believes Gemini currently outperforms its own models as a foundation for Siri, according to an in-depth internal evaluation confirmed this Monday.
A tightly controlled partnership: Google provides Gemini and its cloud infrastructure; Apple leads the interface, personalization, and hardware integration across its devices.
An upgraded Siri: The new version, expected sometime in 2026, will better understand personal context and chain complex actions more smoothly.
A race against time: Delays around Siri had become a serious liability. Gemini offers a way to get back on a roadmap that had turned critical.
Gemini 3 as the trigger: Launched at the end of 2024, the model impressed on benchmarks, delivering the robustness and rapid evolution Apple was looking for.
A change at the top: John Giannandrea left Apple at the end of 2025. Mike Rockwell, formerly of Vision Pro, now leads the AI strategy.
A calculated opening: Apple is also in talks with OpenAI, Anthropic, and Perplexity, avoiding dependence on a single player
Why it's important : Google places its AI at the core of hundreds of millions of devices without ever appearing on screen. Apple, pragmatically, sacrifices technological pride to preserve the experience. Siri changes engines; the ecosystem stays intact.
2️⃣ OpenAI Bets on a Wearable for the Post-Smartphone Era
The summary : A leak attributed to Smart Pikachu reveals an internal OpenAI project codenamed Sweetpea. Behind the understated name sits an audio wearable positioned as a direct rival to AirPods. Sweetpea’s ambition is massive: a potential launch around September 2026 and estimated volumes of 40 to 50 million units. But the surprise isn’t about microphones or sound.
Details :
An always-on assistant: Sweetpea aims to be worn continuously, able to perceive its environment and intervene intelligently, without a screen.
An “earbud” that doesn’t go in the ear: The device would sit behind the ear, freeing up space for electronics and, above all, a larger battery. Enough autonomy for a permanently active assistant.
Industrial ambition that makes your head spin: The leak mentions a launch around September 2026 and production of 40 to 50 million units, a colossal bet for a first hardware product.
The real break is context: According to Wareable, Sweetpea would integrate an ultrasonic transmitter and sensors capable of capturing environmental signals, enabling fine-grained situational awareness.
A rival to the smartphone more than to AirPods: By removing the screen as the entry point, Sweetpea directly challenges the smartphone’s central role in daily usage.
A risky and socially sensitive bet: Costs close to a smartphone, privacy concerns, and acceptance of an “always-on” assistant could slow adoption.
Why it's important : Sweetpea doesn’t promise just another gadget, but a new digital reflex. If OpenAI succeeds, the main interface for AI could leave the pocket and settle discreetly, and durably, behind the ear.
3️⃣ Free AI Is Reaching Its End: OpenAI Moves to Advertising
The summary : OpenAI is preparing to introduce advertising into ChatGPT. In the coming weeks, logged-in adult users on the free and Go plans in the United States will start seeing sponsored content. Ads are clearly separated from chatbot responses. The move is an explicit strategy to finance an AI that is increasingly expensive to run. Plus, Pro, Business, and Enterprise subscriptions will remain ad-free.
Details :
Ads, but at a respectful distance: Advertisements will appear below conversations and will not influence generated answers. The algorithm remains driven by usefulness.
A paid exit ramp: OpenAI guarantees an ad-free experience through its Plus and Pro subscriptions. Users will also be able to dismiss an ad and explain why it’s irrelevant.
Protected topics: No ads will appear in conversations related to health, mental state, or current politics. Teenagers under 18 will also be excluded from targeting.
Privacy under lock and key: Personal data will neither be sold nor shared with advertisers. Users can disable ad personalization and delete data used for targeting.
ChatGPT Go arrives in the U.S.: Launched in India last August, the Go plan is coming to the United States at $8 per month. It offers usage limits multiplied by ten, extended memory, and better context handling, but with ads included.
Tests already underway: According to multiple reports, OpenAI has been experimenting with these ad formats since late 2025, as training and inference costs continue to explode.
Why it's important : With this decision, OpenAI normalizes a future where free chatbots align with the traditional web: useful, powerful… and funded by advertising. A quiet shift, but a structural one for the consumer AI economy.
4️⃣ Simulating 20 Billion Neurons Finally Becomes Possible
The summary : In 2024, the brain of a fruit fly was mapped, revealing 54.5 million synapses and 150 meters of connections. Today, in Jülich, German researchers aim to simulate an entire human brain. They are using one of the most powerful supercomputers on the planet, enabled by record-breaking compute power and large-scale neuronal models.
Details :
From fly to human: Mapping the fruit fly brain in 2024 served as a striking proof of concept, highlighting the extreme complexity of neural circuits even at very small scales.
The return of an abandoned dream: Ten years after the failure of the Human Brain Project, the Jülich research center believes the technical conditions are finally in place to move forward.
JUPITER enters the stage: The simulation relies on JUPITER, ranked fourth worldwide and equipped with thousands of GPUs.Mind-bending numbers: The team led by Markus Diesmann, professor of neurophysics, has already run on JUPITER a spiking neural network equivalent to 20 billion neurons and 100 trillion connections, roughly the scale of the human cortex.
Acknowledged limits: Thomas Nowotny, professor at the University of Sussex, reminds us that even a massive simulation will never fully reproduce the brain’s real functioning.
Why it's important : This breakthrough marks a turning point for computational neuroscience: simulating a human brain could unlock neural dynamics that have so far remained inaccessible. But it also underlines an unsettling truth: even armed with exascale computing, science is still only brushing the surface of the densest mystery in the human organism.
5️⃣ Xiaomi Proves AI Can Also Stay Invisible
The summary : Xiaomi continues to expand its ecosystem and officially launches its first audio smart glasses in France: the Mijia Smart Audio Glasses. They go on sale in Europe at €179. The idea is simple: replace Bluetooth earbuds without turning you into a sci-fi cyborg. No screen, no camera, just sound, comfort, and smooth integration with Android.
Details :
A deliberate featherweight design: The lightest model weighs just 27.6 g (without lenses), barely more than a classic frame and far lighter than the old Bose Frames at 45 g.
Smart, modular temples: All the tech lives in the arms. The Pilot Style and Browline versions feature detachable arms with a proprietary hinge, allowing you to switch between prescription and sunglasses frames.
A more premium variant: The Titanium Alloy model uses a monobloc structure that’s more elegant and robust, but without modularity.
Sound without oversharing: Air conduction via integrated speakers relies on an inverted sound field to limit audio leakage in public spaces.
Well-polished connectivity: Bluetooth 5.4 (on the Titanium version), Google Fast Pair for quick pairing, and Multipoint to switch between smartphone and computer.
Solid battery life: Xiaomi claims up to 13 hours of continuous listening and a full day of mixed use. Charging is handled via a proprietary magnetic connector.
Built-in touch controls: 30 mm touch zones on the temples manage volume, calls, and the voice assistant.
Why it's important : With these audio glasses, Xiaomi takes a pragmatic route: making everyday life smoother without chasing tech spectacle. By focusing on comfort, discretion, and price, the brand helps normalize an object still seen as a gadget and lays a credible first stone on the European smart glasses market.
❤️ Tool of the Week : ChatGPT Translate, AI-powered translation
OpenAI launches ChatGPT Translate, a standalone translation tool accessible via the web. The goal is clear: offer an alternative to Google Translate that’s less mechanical, more intelligent, and better suited to professional and editorial use.
What is it for ?
Translating texts into more than 50 languages, quickly and cleanly
Adapting the tone of a translation (business, formal, neutral, etc.)
Saving time on professional content such as emails, documents, articles, or briefs
Avoiding literal, context-free translations
How to use it
Just head to the ChatGPT Translate website. Paste your text, choose the target language, adjust the style if needed (“more formal,” “more business”), and let the AI do the work. Simple, fast, effective.
💙 Video of the Week : a humanoid that truly acts on its own
LimX Dynamics has released a striking demonstration of its humanoid robot Oli, powered by COSA, an operating system designed for the real world.
In the video, the robot crosses sand, rocks, planks, and debris without losing its balance. But the most impressive part isn’t the walking. It’s what happens before it even takes a step.
Oli receives a spoken instruction, understands the objective, plans its action, adapts its movements, and corrects its decisions in real time, with no preprogrammed trajectory and no teleoperation. It doesn’t follow a script. It reasons while moving.
This demo marks a clear break from classical robotics. Here, perception, language, decision-making, and movement are no longer separate blocks. Everything is integrated into a single architecture inspired by how the human brain works.
You’re no longer watching a robot “doing a demo.” You’re observing an embodied agent capable of operating in unpredictable environments. Less spectacular than a dancing robot, but infinitely closer to a humanoid that’s genuinely useful.
Has Apple lost the AI war?
