Realistic robots, a supercharged Siri, an insomniac ChatGPT... who needs humans anymore?

👋 Dear Dancing Queens and Super Troupers,

Picture this: Mark Zuckerberg adjusts his connected Ray-Bans, looks up from his screen, and coldly declares: “The future is humanoid.”
No, it’s not a scene from Black Mirror — it’s Meta’s latest pivot. After pouring billions into the metaverse, the company is now launching Metabot, an “AR-sized” bet that could reshape our homes, workplaces… and maybe our very lives.

But careful: the project isn’t about building robots in factories — it’s about providing the brains (powered by Llama) to Tesla Optimus, Figure 02, and Unitree G1. In short, turning Meta into the Android of humanoids.

While Meta prepares for the robot wars, OpenAI has unveiled its massive measuring weapon: GDPval, a benchmark comparing GPT-5 (and Claude Opus 4.1) to real professionals across 44 jobs — from software engineers to nurses.
In 40–50% of cases, the AI was judged as good as — or even better than — the experts. Enough to send chills down some spines… and give others a boost in productivity superpowers.

Meanwhile, Google DeepMind is moving its robots from the “paper folding” level to “packing a suitcase based on London weather.”
Thanks to Gemini Robotics 1.5 and ER 1.5, machines can now chain multi-step tasks, collaborate across robots, and even pull data from Google Search to sort your recycling according to local rules. Yes, this is real.

And as if that weren’t enough, here comes ChatGPT Pulse. The idea? Let AI work while you snore. Every morning, it drops a personalized bundle of briefs — news, schedule, sorted emails, travel ideas — straight into the app.
For now, it’s reserved for Pro subscribers at $200/month, but it’s a glimpse of the personal assistant that could replace your news feed.

Not to be left behind, Apple is quietly testing Veritas — an internal project that fuses Apple AI with Gemini to create a super-charged Siri, able to act inside apps and respond in context. The catch? It may never leave Cupertino’s walls.

Notice a pattern? The GAFAM are going all-in on AI, giving us a preview of just how central this technology is set to become in the near future.

Here’s this week’s lineup :

👉️ Meta’s new obsession: humanoids powered by Llama ​​🤖​​​​​​​​​​​​​​​​​

👉️ OpenAI pits 44 jobs against AI: who keeps their work? 🕴️​

👉️ Google teaches robots to fold shirts and pack suitcases ​🧳

👉️ AI wakes up earlier than you: introducing ChatGPT Pulse 🧐​​

👉️ Siri 2.0: Apple’s secret upgrade… that might never launch 📱​​

If we've forwarded this letter to you, subscribe by clicking on this link !

🔥 If you have 15 minutes

1️⃣ Meta’s new obsession: humanoids powered by Llama

The summary: Meta just won’t quit. After burning billions on its pixelated mirage of a metaverse, Mark Zuckerberg is pulling another trick from his hat: humanoid robotics. The project, dubbed Metabot, comes with a jaw-dropping $65 billion investment.

The twist? No titanium arms or Tesla-style supersonic knees. Meta is betting on the brain and the software. The idea is to provide its platform to robot makers that meet its standards. For now, Metabot looks more like a brilliant draft than a suited-up Terminator. But patience — the creature is still in its infancy.

Details :

• A $65B ticket : Reality Labs, already busy inflating AR, gets a robotics division. The goal: tie robotics to Meta’s headsets and glasses.

• A star-studded cast : Marc Whitten (ex-Microsoft, Amazon, Sonos, Unity) leads the project, alongside MIT’s Sangbae Kim, hailed as “the best tactical robot builder.”

• Software before bolts : CMO Andrew Bosworth insists the real bottleneck is code. Hardware matters, but it’s secondary. The ambition? Build a universal humanoid platform — think Android for robots.

• Superintelligence Labs as backup : Meta leans on its MSL, launched in June. First mission: teach a robotic hand the dexterity of a pianist.

• Weaker than Tesla’s Optimus : Musk’s toy boasts 23 joints. Metabot, for now, might settle for two working thumbs.

Why it's important : Instead of competing on mechanical design, Meta wants to become the go-to publisher of robotic brains. If the bet pays off, it won’t sell the best humanoid, but it could become the Android of global robotics. And that’s worth more than two thumbs up.

2️⃣ OpenAI tests 44 jobs against AI: who keeps their work?

The summary: OpenAI introduces GDPval-v0, a benchmark that pits its models against human pros across 9 industries and 44 professions. The numbers? GPT-5-high rivals or matches experts 40.6% of the time, while Claude Opus 4.1 scores an even higher 49%.

For now, the challenge is limited to written reports — not the whole job. Stronger versions are coming. Behind the scenes, Dr. Aaron Chatterji and Tejal Patwardhan highlight the massive jump since GPT-4o (a meager 13.7% just 15 months ago).

Details :

• GDPval, explained : pros review AI-written reports against human ones and pick the winner. Version v0 focuses only on writing.

• Nine pillars of the economy : health, finance, industry, public sector, and other US GDP drivers were put to the test.

• 44 jobs on trial : from software engineers to nurses to journalists, deliverables are judged on quality.

• GPT-5-high score: 40.6% : better than or equal to human experts on 4+ tasks out of 10, with extra computing power.

• Claude 4.1 score: 49% : Anthropic shines, helped by reports judged as “prettier,” says OpenAI.

• Admitted limits : writing a report ≠ doing the job; OpenAI is preparing broader, more interactive benchmarks.

• Execs weigh in : Chatterji says it frees time for higher-value work. Patwardhan points to the leap from GPT-4o’s 13.7%.

Why it's important : If models are already flirting with expert-level outputs on real deliverables, AI becomes a credible teammate. But as long as GDPval is still v0 and text-only, don’t pop the champagne yet — the real game will be full workflows, far beyond polished reports.

3️⃣ Google teaches robots to fold shirts and pack your suitcase

The summary : Google DeepMind has launched Gemini Robotics 1.5 and Gemini Robotics-ER 1.5: a duo that lets robots plan multi-step tasks. They can search the web when stuck, then act in the real world.

Carolina Parada announced the shift from single-command demos to full-on problem solving. Bonus: skill transfer between robots and developer access via the Gemini API in Google AI Studio (ER 1.5), while Robotics 1.5 remains for selected partners. Previous versions dropped in March.

Details :

• From folding to reasoning : No more tiny demos like folding paper or unzipping a bag. Now it’s laundry sorting, suitcase packing based on London weather, and recycling adapted to local rules.

• A dual-brain system : Robotics-ER 1.5 interprets the environment and taps into Google Search. It then translates that into natural instructions for Robotics 1.5, which executes step by step.

• Robot-to-robot knowledge transfer : Skills learned on ALOHA2 (a dual-arm robot) transfer to Franka (another bimanual bot) and even to Apptronik’s humanoid Apollo. Proof, says Kanishka Rao, that one model can pilot very different robots.

• Access & rollout : Robotics-ER 1.5 arrives via API for all developers; Robotics 1.5 stays limited to a few hand-picked partners.

• Scaling up : Parada emphasizes the leap from “one command at a time” to contextual understanding with an action plan.

Why it's important : Robots that search, understand, and transfer skills are edging closer to becoming all-purpose assistants — the kind that can unpack an apartment without ever checking the manual. If developer access delivers, robotics could speed up dramatically, from lab to laundry room.

4️⃣ AI wakes up earlier than you: here’s ChatGPT Pulse

The summary: OpenAI launched ChatGPT Pulse last Thursday. Pulse prepares overnight morning briefings that appear when you wake up, formatted as 5–10 cards (text + images) in a new tab.

Initially available for Pro subscribers at $200/month, the tool is pitched as a proactive assistant “for everyone,” according to Fidji Simo. But broader rollout depends on efficiency and server capacity, backed by Oracle and SoftBank. Sam Altman reminds us that heavy-duty features will remain premium at first.

Details :

• Tailored briefings : From Arsenal match updates to travel routes to Sedona and Halloween costume ideas, demoed live by Adam Fry.

• Interactive cards : You can open the full card, ask questions, and request fresh automated reports.

• No more doomscrolling : A final “Great, that’s all for today” message cuts the endless scroll.

• Plugged into your tools : Google Calendar and Gmail connectors allow overnight analysis of emails and agenda, served up as a ready-to-go morning plan.

• Useful memory : With memory enabled, Christina Wadsworth Kaplan showed how Pulse picks up on personal context — like a love for running or a pescatarian diet.

• Roadmap ahead : A more agentic Pulse is coming — think restaurant bookings or draft emails for approval. But that part isn’t ready yet.

Why it's important: Pulse shifts ChatGPT from reactive answers to proactive initiative : a companion that watches, filters, and synthesizes while you recharge. If rollout expands, newsletters, news apps, and aggregators may feel the wake-up call — no snooze button allowed.

5️⃣ Siri 2.0: Apple tests a supercharged version… that may never see the light of day

The summary: Mark Gurman (Power On) is urging Apple to bring its internal “ChatGPT-style” app into the open to give credibility to its AI pivot. In the same newsletter edition: new MacBooks and Apple monitors lining up for launch, an iPhone 17 Pro “scratchgate” hurting the premium image, and a fresh memo from Tim Cook to re-energize the troops.

Details :

• A bot worth the spotlight : Apple has a ChatGPT-like conversational app, already tested internally. Publishing it would strengthen confidence in its AI strategy.

• Macs warming up backstage : New MacBooks are moving in step, alongside Apple-branded external displays. Both are inching closer to launch day.

• “Scratchgate” hits the iPhone 17 Pro : Complaints are mounting over premature micro-scratches on the chassis and finish, denting the “premium” promise.

• Word from the boss : Tim Cook issued a fresh memo, using rhythmic internal messaging to keep teams aligned with the product roadmap and Apple’s homegrown AI push.

• Power On, Gurman-style : All of this comes from Mark Gurman’s Bloomberg newsletter (Power On), the unofficial compass for Apple launches (dated Sept 28, 2025).

Why it's important: Apple’s credibility in its AI pivot is on the line: releasing the chatbot app publicly would prove Cupertino is delivering more than slogans. If MacBooks and monitors launch as a duo, the ecosystem regains momentum — provided Apple douses the “scratchgate” flames before memes make more noise than the keynote. In short: a week where AI wants the stage, hardware lines up its entrance, and Tim Cook holds the baton.

❤️ Tool of the Week: Suno V5, l’IA qui compose vos hits

Suno V5 is the new version of the viral AI music generator — more realistic, more emotional, and more versatile than ever. It can create full songs (vocals + instruments + lyrics) from a simple prompt, whether you’re a pro musician or just curious to test wild ideas.

What’s it for?

• Compose in any genre : ballad, electronic, educational rap, or even a quirky cartoon theme.

• Choose your voices : male or female, tailored to the mood.

• Dial in the emotion : melancholy, intensity, energy — all serving your storytelling.

• Produce coherent tracks : lyrics and melodies finally woven seamlessly.

• Save time : intuitive interface, no technical skills needed.

• Explore creative projects : from pro use (films, podcasts, content) to fun experiments (parodies, side projects).

How to use it?

• Go to the Suno AI platform (subscription required).

• Enter a prompt describing th

• e style, genre, and mood you want.

• Pick the voice and emotional intensity.

• Let it run… and in seconds, your track is ready to play.

💙 Video of the Week : Only Head, the hyper-realistic robot that chills the blood