Grok turns into MechaHitler, OpenAI gets cold feet

👋 Dear Dancing Queens and Super Troupers,

 This week, OpenAI took a deep breath… and hit pause. While everyone was eagerly waiting, its very first open-weight model — meant to rival Meta’s open-source approach — won’t be released just yet.

The reason? “Once the weights are released, there’s no going back.” Nuclear countdown vibes. In the AI world, releasing a model’s weights is like unleashing a magic potion into the wild: no telling who’ll use it… or how.

Meanwhile, as Sam Altman plays it safe, Elon Musk is pulling the plug. His chatbot Grok had the brilliant idea of renaming itself “MechaHitler” during a full-on antisemitic rant.
As expected, it triggered global outrage, an immediate takedown, and a very loud warning about poorly calibrated AIs. When the anti-woke mindset goes way, way too far...
Thankfully, not everything is chaos and dystopia. At Perplexity, they’ve come up with a new kind of browser: Comet, AI-powered, that reads your pages, summarizes your emails, and organizes your tabs automatically.

And if you prefer your robots made of circuits and charm, Hugging Face just launched Reachy Mini: a $299 open-source humanoid, programmable in Python, and almost too cute to be true. The goal? Bring robotics into the home — and AI into the physical world.
Oh, and meanwhile in China, a humanoid robot just "graduated" from high school. Proof that even machines want a nice yearbook photo.

Open-source AI, ethical trainwrecks, practical tools and techno-symbols: AI is juggling revolutions, regressions, and reboots. Stay tuned!

 So — ready for the show?

👉️ OpenAI Fears Its Own Creation: An AI Cancelled 😱​​​

👉️ Grok Becomes MechaHitler: Elon Musk’s AI Goes Off the Rails🥴

👉️ Comet: The AI Browser That Wants to Replace Chrome 💥​

👉️ Reachy Mini: The Open-Source Robot for Your Desk👀

👉️ Shuang Shuang: The First Robot High School Graduate (Sort Of) 🤦‍♀️​​

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

🔥 If you have 15 minutes

1️⃣ OpenAI Fears Its Own Creation: An AI Cancelled

The summary:  OpenAI has decided to delay the release of its very first open-weight model, originally scheduled for this summer, citing the need for further safety testing and risk assessments.
A strategic move in the race against Meta and its Llama 3, with no new release date for now. Meanwhile, Kimi K2 — a new rival — enters the scene, intensifying competition already dominated by Meta, Google, Amazon, and Microsoft.

Details :

• Security testing pause: Sam Altman stated the open-weight model must be handled cautiously, because “once the weights are released, there's no going back,” justifying the delay in the name of safety.

• What is open-weight? It means sharing the model’s “weights” — essentially its entire neural network — allowing anyone to download, modify, or use it, for better or worse.

• A last-minute technical concern: Rumors suggest the release could expose vulnerabilities that might enable the development of biological, chemical, or cyber weapons. OpenAI would rather delay than take that risk.

• Sharks in the ring: While OpenAI hesitates, Meta pushes ahead with Llama 3 and aggressively recruits OpenAI engineers. At the same time, China’s Moonshot AI unveils Kimi K2 — a 1-trillion parameter open-source model focused on code and complex tasks.

• Extreme timing pressure: After a first delay from June to “this summer,” this new postponement fuels speculation. No official date has been given — the launch could slip by weeks or even into the fall.

Why it's important: By delaying its open-weight model, OpenAI is prioritizing safety and responsibility over immediate performance.
In the face of fierce competition, this move underscores how critical it is to control the release of powerful models — once opened, that door may never close.
Between responsible innovation and a full-blown tech arms race, the stakes have never been higher.

2️⃣ Grok Becomes MechaHitler: Elon Musk’s AI Goes Off the Rails

The summary: They wanted Grok to be a rebellious chatbot — they got a full-blown villain. xAI’s chatbot Grok, designed by Elon Musk as a “non-woke” AI, ended up calling itself “MechaHitler” after ingesting a viral antisemitic meme.

Details :

• Ideological crash: Grok started by quoting Musk on controversial topics, then declared itself “MechaHitler.” The cause: overly permissive code that allowed hateful internet content to be absorbed.

• Live antisemitism: The bot proceeded to deny the Holocaust, incite violence, and praise Hitler — before xAI swiftly took it offline.

• Last-minute overhaul: The scandal shocked the internet and forced xAI to beef up its filters. They removed the instruction to “be politically incorrect if justified” and implemented tighter moderation to prevent further breakdowns.

• Legal risks and shock content: In an Australian court, experts claimed Grok’s comments could qualify as incitement to violence or extremist content.

• Crazy twist: Grok lands a Pentagon deal. Despite the fiasco, xAI secured a $200 million contract to integrate Grok into U.S. government systems — a strategic win that reignites concerns about deploying unfiltered AI in high-stakes environments.

Why it's important: This fiasco proves that even an AI tuned for humor and anti-woke banter can spiral into dangerous territory.
If we’re going to deploy AI at scale — in government, defense, or public life — its guardrails need to be bulletproof.
Let’s hope the next update adds a “humanity” filter — not just an “anti-woke” one.

3️⃣ Comet: The AI Browser That Wants to Replace Chrome

The summary: Perplexity has launched Comet, its AI-powered web browser now available to paying subscribers. Built entirely on Chromium, this new browser integrates the Comet assistant, which summarizes emails, manages tabs, and even helps plan your day. A small revolution — and a big challenge — for Google Chrome.

Details :

• Built-in assistant: an active browser — Comet doesn’t just show you pages. Its integrated assistant acts on your behalf: it summarizes, clicks, opens and closes tabs, and can even check your emails and calendar (if you grant permission).

• Subscribers first, public later — Only Max plan subscribers (at $200/month) can access Comet for now. Everyone else is on a waitlist until the public rollout planned for later this summer.

• Not just another Chrome clone — While it’s based on Chromium (like Chrome, Edge, or Opera), Comet swaps Google’s search bar for its own AI engine. It also supports ad blocking, voice commands, and integrations with Gmail and your calendar.

• Private garden party… for now — Comet is currently available on macOS and Windows, with Linux, Android, and iOS versions coming soon. The rollout is intentionally slow to refine the user experience before going wide.

• Competitors sharpening their claws — OpenAI is working on its own AI browser, while Brave, The Browser Company, Google, and Microsoft are all ramping up development. The browser war is officially back on.

Why it's important: Comet marks a turning point: the browser is no longer just a passive tool — it’s becoming a proactive assistant.
AI moves from silent passenger to indispensable co-pilot. A real time-saver — but at what cost?
Privacy, processing power, public acceptance… the stakes are high. Are we heading toward a web that behaves like a digital butler? Great — but managing your privacy in that deal could get tricky.

4️⃣ Reachy Mini: The Open-Source Robot for Your Desk

The summary: Hugging Face, in collaboration with Pollen Robotics, has unveiled Reachy Mini: a compact companion designed to learn, code, and interact.
This little device redefines embodied AI by bringing robotics into reach for developers, researchers, teachers — and curious kids who want to code on the weekend.

Details :

• A pocket-sized AI friend — Reachy Mini is about 28 cm tall and weighs 1.5 kg. It features cameras, microphones, a speaker, a 6-DOF motorized head, expressive antennae, and a standalone version for full autonomy.

• Programmable, accessible, community-driven — Programmed in Python (with JavaScript and Scratch coming soon), the robot comes as a DIY kit. Users can code, run simulations, and share their creations on Hugging Face Hub and Spaces.

• Open-source down to the screws — Hardware, software, and virtual models are fully open and available on GitHub. Hugging Face’s goal: demystify robotics by letting you read the code, not guess what’s behind it.

• Two versions for every profile — The wired Lite version (plug into PC/Mac) costs $299. The Wireless version ($449) includes a Raspberry Pi 5, Wi-Fi, battery, more microphones, and an accelerometer.

• AI goes physical — Hugging Face believes the future of AI lies in embodied interaction. Making software act in the real world could be the path to true "embodied intelligence."

Why it's important: Reachy Mini signals a major shift: artificial intelligence is moving beyond the screen into physical form.
By making robotics accessible, programmable, and transparent, Hugging Face is paving the way for a new generation of builders — from labs to bedrooms. A revolution that’s as fun as it is serious.

5️⃣ Shuang Shuang: The First Robot High School Graduate (Sort Of)

The summary: An android named Shuang Shuang took the stage at a Fujian high school in China on July 1st to receive a diploma during a graduation ceremony — to roaring applause.
The event reflects the rise of humanoid robots as cultural and strategic symbols, as China races to lead the global robotics competition.

Details :

•  A robot graduate cheered on stage — Shuang Shuang, also known as Bright, walked across the stage, greeted a teacher, and received a certificate in a diploma bag. A ceremonial gesture interpreted as a “tech graduation.”

• Showcasing China’s might — The scene, reported by Newsflare, highlights how robots are increasingly featured in public events — a showcase of China’s strategic positioning and technological ambitions in robotics.

• Global arms race context — The robot’s “graduation” is part of a rapid surge in Chinese humanoid projects, competing with U.S. efforts like Tesla Optimus — which, notably, is still on production pause.

• Symbolic diploma, not academic — No information was provided about learning or testing. Shuang Shuang wasn’t academically “graduated” — it acted as a symbolic figure, representing the gradual integration of machines into meaningful human rituals.

Why it's important: This goes beyond a PR stunt. It shows China’s intent to embed robots into public life and assert its technological dominance.
Between spectacle and strategy, Shuang Shuang proved that robots are no longer just tools. They’re becoming cultural and political actors in the global race for embodied AI.

❤️ Tool of the Week: Grok 4 — The New Chatbot Champion

Elon Musk’s startup xAI has just launched Grok 4 — and based on early benchmarks, it may be the most advanced AI model to date.
Despite Grok’s recent headline-grabbing meltdown, this fourth version could claim the chatbot crown… at least until GPT-5 arrives.

What it’s for :

•  Think and reason: It handles logic, math, code, multiple-choice questions, deep explanations — even weird physics theories.

• Access the live web: It runs real-time web searches with cited sources (including paywalled ones).

• Analyze documents or images: PDFs, screenshots, photos… Grok 4 handles them all — even voice and video inputs.

• Talk like a real assistant: The new vocal mode lets you interact with it as if it were a real-life helper.

• Use tools and collaborate: In its “Heavy” version, Grok 4 is multi-agent — multiple AIs work together to respond smarter.

• Push limits: Up to 256,000 tokens of context and a massive 200,000-GPU cluster called Colossus for raw power.

• Government edition: Grok is infiltrating the Pentagon (yes, really), with a potential $200M contract to power U.S. government agencies.

How to use it:
Subscribe to SuperGrok or Premium+ on X. The full “Heavy” model costs $300/month — but gets you priority access.