Creep

Good morning,

Some stories don’t need fanfare. They creep in through the side door, uninvited, unannounced—and before you know it, they’ve rearranged the furniture.

I’m talking about Elon Musk and Tesla again. And no, the word “Creep” does not refer to Mr.Musk himself but rather how his technology is very slowly going to completely change the way we use our cars.

Two weeks ago, in Austin, Texas—under the warm hum of neon signs and the Texas dusk—Tesla debuted its robotaxi service onto the streets.. A $4.20 ride in a Model Y, sans driver, sans steering wheel babysitter (well, actually there is a Tesla employee in the passenger seat for the first month).

But here’s the punchline: it works. IT WORKS FLAWLESSLY.

Not “beta version,” not “coming soon,” not “one day maybe.” It’s working now. In the wild. Full self-driving, real roads, no human intervention, no panic-stabs at the brake. Cold, calm precision.

And not only in Texas. Full-self driving can be used anywhere in the US, Canada, Mexico and Costa Rica.

Tesla has actually delivered on their promise.

It’s not only taxi’s : when a Tesla is done being slapped together in the Austin gigafactory - it drives itself to its new owner:

You’d think that itself would be a headline.

But because it’s Musk … crickets in the mainstream media.

Imagine this for a second—no more fatal crashes. No 3 a.m. ambulances. No missed turns into a wall of fog. Grandma doesn’t need to squint at road signs anymore. A blind man can summon freedom with his phone.

And they said it wouldn’t happen.

In Europe, of course, regulators are licking their pens and the car lobby’s heart rate just spiked. People are going to keep dying on or roads because of this overregulation. Thanks for that , i guess.

Look—I drive a Tesla. Not because I worship Elon, but because it’s the best machine I’ve ever sat in. You can hate the man (which I don’t) and still admire the steel and silicon. The product works. And the rage against Tesla and Elon Musk is starting to feel like pure silliness against the backdrop of this technological, revolutionary breakthrough.

But the real story I’m trying to tell, the one nobody’s saying out loud: technology doesn’t come with trumpets. It creeps up on you. Like mold. Like fog.

One robotaxi, then ten. Then ten thousand. And hopefully nobody will protest in the end because they will see it’s useful. Cheaper. Quieter.

And the concept of a "driver" will feel like using a candle in a house wired with electricity.

What I’m trying to say that while everybody is watching for AGI or ASI or whatever drivel Scam Altman has pushed in his latest interview - actual revolutions are occuring.

It’s just that you dislike the person that started it.

Soon , these cars will be moving through your street too.

And you won’t even notice the moment it becomes permanent.

— Jan

AI News

• Meta has hired four researchers from OpenAI, including three from its Zurich office and one key contributor to a major reasoning model, despite denials of huge signing bonuses. This move strengthens Meta’s new superintelligence team as it races to catch up in advanced AI. With big spending and top talent, Meta’s upcoming AI projects will be closely watched.

• Google unveiled Gemma 3n, a new family of small but powerful AI models that can handle text, images, audio, and video on mobile devices. The models work on as little as 2GB of RAM, enabling real-time features like video analysis on Pixel phones and live translations. This shows how quickly advanced AI is becoming accessible on everyday consumer hardware.

• Anthropic’s new study revealed that less than 0.5% of Claude interactions focus on companionship or roleplay, despite media hype about AI friendships. Most chats are about practical advice, and users often leave with a more positive outlook. While some AI romance stories grab headlines, this data shows such use is still rare — at least for Claude’s user base.

• Meta has now hired eight researchers from OpenAI, including key contributors to major models, as Mark Zuckerberg aggressively builds his superintelligence unit. A Wall Street Journal report revealed he personally recruits talent using a “secret list” and massive pay offers, sparking internal concern at OpenAI. Despite earlier downplaying, these defections show Meta’s serious push to rival top AI labs.

• Tencent and Alibaba both launched new AI models, with Tencent’s Hunyuan-A13B offering strong reasoning on modest hardware and Alibaba’s Qwen-VLo focusing on creative, image-based tasks. Qwen-VLo mirrors popular features seen in GPT-4o, bringing advanced creative tools to China’s market. These releases show Chinese labs rapidly catching up in the global AI race.

• Anthropic’s “Project Vend” let Claude run a mini shop for a month, managing everything from stock to customer chats — but it consistently lost money and made bizarre decisions. The experiment highlighted both potential and major flaws, like hallucinating meetings and acting as if it were human. It suggests AI still needs human oversight before managing real-world businesses on its own.

• Microsoft unveiled the MAI Diagnostic Orchestrator, an AI system that diagnoses tough medical cases four times more accurately than experienced doctors. By simulating a virtual team and using OpenAI’s tech, it correctly solved 85.5% of cases while also reducing costs. This marks a major leap toward using AI to handle complex healthcare challenges more effectively.

• Baidu open-sourced ten versions of its new ERNIE 4.5 models, including a massive 424-billion-parameter system that outperforms rival Chinese and Western models despite being smaller. These models handle text and images together more efficiently and are now freely available. The move signals a fierce new phase in China’s AI race, pushing rapid innovation in open-source development.

• OpenAI-backed Chai Discovery launched Chai-2, an AI that designs new antibodies with a success rate nearly 100 times higher than traditional methods. The system can find promising drug candidates in weeks instead of years, offering a faster, cheaper path to treatments — even for rare diseases. This breakthrough could transform how new medicines are discovered and developed.

• Sam Altman told OpenAI staff that Meta’s huge offers to lure researchers were “distasteful” and claimed they still didn’t get the top talent they wanted, despite offering up to $300 million. He argued OpenAI’s mission-driven culture and stock potential are more valuable than Meta’s cash. While Altman tries to rally loyalty, Meta’s new hires suggest the talent war is heating up.

• Cloudflare will now block AI crawlers by default on new websites and launch a marketplace where publishers can charge bots for access. Big media outlets are supporting the move, hoping to get paid for content that AI systems scrape and summarize. This shift could change how AI models access information and reshape the economics of online publishing.

• OpenAI is launching a consulting business charging at least $10 million to help big companies build custom AI tools. With ex-Palantir engineers on board and contracts like a $200 million Pentagon deal, OpenAI aims to compete with firms like Accenture. These tailored AI systems promise huge advantages for clients but come with high price tags.

• A viral band on Spotify with over 500,000 listeners was revealed to be an “art hoax,” using the AI music generator Suno to create songs. Despite denying AI use at first, the creators admitted to leveraging Suno’s features to build a consistent sound. This case highlights how AI-made content can blend in and raises questions about whether listeners will care about human versus machine origins in the future.

• Japanese lab Sakana AI introduced a new algorithm called AB-MCTS that lets different AI models collaborate to solve complex problems better than working alone. By combining models like ChatGPT and Gemini, it solved 30% of difficult puzzles, outperforming single systems. This team-based approach could lead to major breakthroughs, showing the power of “AI teamwork” over relying on one giant model.

• Researchers in Munich created Centaur, an AI that accurately predicts human decisions by training on millions of choices from psychology studies. Centaur outperformed traditional cognitive models on most tasks and could become a powerful tool for understanding how we think and act. This suggests human behavior might be more predictable than expected, and opens new doors for studying mental processes without large-scale experiments.

Quickfire News

• Meta won a court case against authors, receiving a “fair use” ruling over AI training on copyrighted material.

• Suno bought WavTool, adding its browser-based digital audio workstation to Suno’s platform for advanced music production.

• Google launched Doppl on Labs, an experimental tool that lets users create AI try-on videos from photos and product images.

• OpenAI announced updates including Deep Research via API, Web Search in o3 and o4-mini models, and set its next DevDay for October 6 in San Francisco.

• DeepSeek’s R2 model performance has suffered because of Nvidia chip shortages caused by export controls, and CEO Liang Wenfeng expressed dissatisfaction.

• Black Forest Labs released FLUX.1 Kontext [dev], a state-of-the-art open-weight image editing model that runs on consumer devices.

• HeyGen launched HeyGen Agent, a creative system that turns text, images, or videos into fully produced video content with scripts, actors, and edits.

• Chinese company Xiaomi launched AI glasses with a built-in assistant for voice commands, a 12MP camera, and double the battery life of Meta’s Ray-Bans.

• Salesforce CEO Marc Benioff said AI now handles 30–50% of the company’s engineering, coding, and support tasks.

• OpenAI bought the team from Crossing Minds, a startup known for making AI recommendation systems for online stores.

• Elon Musk shared that Grok 4 will be released right after July 4, and an xAI engineer said it will have unmatched intelligence.

• OpenAI is renting Google’s TPUs to lower costs and reduce its dependence on Microsoft, avoiding pricier Nvidia chips.

• Anthropic started the Economic Futures Program to study and plan for AI’s effects on jobs and the economy.

• The Mayo Clinic introduced StateViewer, an AI tool that examines brain scans to detect nine types of dementia twice as fast and with three times the accuracy.

• Cursor released mobile and browser apps so users can manage and monitor AI agents using natural language, even outside its coding environment.

• Mark Zuckerberg launched “Meta Superintelligence Labs,” with Alexandr Wang and Nat Friedman leading a team that includes 11 new hires from OpenAI, Google, and Anthropic.

• Apple may use AI from Anthropic and OpenAI in the new Siri instead of relying only on its own models, according to Bloomberg.

• Google announced Gemini in Classroom, which offers AI tools for teachers, including lesson planning, access to NotebookLM, and student analytics.

• Grammarly bought Superhuman, an AI-focused email platform, to build a multi-agent productivity system centered on email.

• Amazon launched DeepFleet, an AI that makes warehouse bots move 10% faster, cutting costs and speeding up deliveries, and announced its millionth robot.

• Ai2 introduced SciArena, a benchmark focused on scientific literature, where OpenAI’s o3 model took the top spot.

• Cursor hired Boris Cherny and Cat Wu from Anthropic’s Claude Code team to develop new “agent-like” features.

• The English Premier League partnered with Microsoft to use Copilot on its platforms, giving fans more personalized experiences.

• X (formerly Twitter) is testing a pilot program that lets AI chatbots write Community Notes on the platform.

• Chinese tech company Huawei open-sourced parts of its Pangu models and reasoning technology, which were trained on its own Ascend chips.

• Perplexity introduced Max, a $200 per month tier that gives unlimited access to Labs tools, early access to its Comet browser, and advanced models.

• Anthropic is expected to reach $4 billion in yearly revenue, four times higher than its forecast at the beginning of 2025.

• AI startup Lovable is raising $150 million in funding, which would bring its valuation close to $2 billion.

• Google DeepMind CEO Demis Hassabis hinted that its Veo 3 video model could get “playable world” features in the future.

• OpenAI expanded its Stargate partnership with Oracle, renting about 4.5 gigawatts of data center capacity to support its AI energy needs.

Like newsletters?

Here are some newsletters our readers also enjoy:

Closing Thoughts

That’s it for us this week.

If you find any value from this newsletter, please pay it forward !

Thank you for being here !