Tesla Robotaxi Rolls Out - Sync #525

Plus: Thinking Machines Lab raises $2B; Gemini Robotics On-Device; Apple in talks with Perplexity; plants genetically engineered to purify air; Anthropic and Meta score wins in copyright lawsuits

Hello and welcome to Sync #525!

After years of promises, Tesla has finally launched its Robotaxi service in Austin, Texas. In this week’s issue of Sync, we take a closer look at how the launch went and how Tesla’s offering fits into the wider robotaxi landscape.

Over in AI, reports have emerged that Thinking Machines Lab—a secretive AI start-up founded by Mira Murati, former CTO of OpenAI—has raised $2 billion at a $10 billion valuation. Meanwhile, Apple has reportedly considered either partnering with or acquiring Perplexity; Anthropic and Meta score wins in copyright lawsuits; and OpenAI must deal with a copyright lawsuit over the “io” trademark.

In robotics, Google has released an on-device version of Gemini Robotics, a German robotics company has announced a new humanoid robot, and Foxconn and Nvidia are reportedly in advanced talks to deploy humanoid robots at a new Foxconn factory in Houston.

Additionally, Google has released Gemma 3n, Imagen 4, and AlphaGenome—a new AI model to predict how genetic variations in human DNA affect gene regulation. This week’s issue of Sync also features genetically engineered plants designed to purify air, a BCI capable of synthesising near-instantaneous speech by decoding neural signals in real time, how AI is changing software (again), and more news from AI, robotics, and beyond!

Enjoy!

Tesla Robotaxi Rolls Out

After years of promises, Tesla has finally launched its Robotaxi service in Austin, Texas. While the company is calling the move a historic milestone, it is not without caveats.

Image created with Midjourney

First, this was a small-scale launch, with just ten to twenty Tesla Model Y vehicles operating within a strictly geofenced area of South Austin. Each ride is monitored by a “safety monitor”—a Tesla employee in the front passenger seat. There is no human behind the wheel, making this the first time Tesla has transported paying passengers without a driver present. However, with a safety monitor on board and the exact limits of their intervention unclear, the service is not fully driverless in the strictest sense.

Not everyone can hail Tesla’s Robotaxi, either. The company has so far avoided opening the service to the general public, instead offering the first rides to carefully chosen early-access users, many of whom are well-known Tesla enthusiasts with large online audiences.

Rides cost a flat fee of $4.20 (because of course). Each trip also comes with a list of restrictions: riders must be over 18, rides are available only between 6 a.m. and midnight, and the robotaxis will not operate in inclement weather or other challenging conditions. The number of passengers per ride is limited to two.

Not So Smooth a First Ride

While Elon Musk declared the launch the “culmination of a decade of hard work”, it has not been a smooth ride. Social media quickly filled with videos from invited riders sharing their experience, some of which showed robotaxis exceeding speed limits, hesitating at intersections, and swerving unexpectedly into the wrong lane. In at least one case, a robotaxi stopped abruptly in the middle of the road after a passenger pressed a button requesting it to pull over. Remote customer support was required to resolve the situation and continue the journey.

As Bloomberg reports, the US National Highway Traffic Safety Administration (NHTSA) has responded by reaching out to Tesla for more information and continues to closely monitor the trial. So far, no formal investigation has been opened, but the agency emphasised that it will “take any necessary actions to protect road safety” as incidents are reviewed.

Tesla’s approach to transparency has also raised eyebrows. The company has resisted efforts by journalists to obtain more details about the service, its deployment plans and technical underpinnings, citing trade secrets and competitive concerns. A request for public records filed by Reuters has been contested by Tesla through legal channels, highlighting the company’s desire to tightly control information around the launch.

In all fairness, it is worth noting that Waymo and other companies are not perfect either. Over the past year or so, there have been numerous instances of Waymo’s robotaxis becoming confused and blocking traffic, requiring intervention by a human teleoperator to resolve the issue.

The Road Ahead

The Austin launch comes at a pivotal time for Tesla. The company is in a tough spot, to say the least. Car sales are declining worldwide, and competition is catching up and, in some cases, like BYD in China, surpassing Tesla. Additionally, thanks to Elon Musk’s involvement in US politics, Tesla’s brand image is tarnished for many potential customers.

With all of this happening, Tesla is trying to carve out a new road ahead, one of which is robotaxi services. This might be a good long-term strategy. Instead of being a car maker that needs to convince customers to buy a new vehicle, Tesla could refocus on robotaxi services. This is a more predictable business model—not everyone will buy a Tesla, but there is a much larger number of people inclined to use a robotaxi.

Waymo is already proving that focusing solely on robotaxi services makes sense. The company leads the robotaxi industry in the US, running over 250,000 paid rides per week and is targeting 10 cities by year-end (Waymo is already operating in Austin). Its 2025 revenue is projected at $181 million to $230 million, with some analysts predicting $6.1 billion in bookings by 2034. Additionally, Waymo uses off-the-shelf cars (at least for now), which brings costs down and allows it to focus solely on self-driving technology.

While Tesla’s ambition is bold, there are reasons to be sceptical. The Austin pilot was highly choreographed: a handful of vehicles operating in a restricted area, with safety monitors on board, and rides reserved for friendly social media influencers. Expanding from this tightly controlled environment to real-world, large-scale deployment is a leap that analysts warn will be “orders of magnitude” more difficult.

Additionally, Tesla will need to prove that its camera-only approach and AI can handle complex edge cases in unpredictable traffic, a problem that has taken rivals like Waymo more than a decade to address using a more cautious, sensor-rich strategy. Some analysts warn that if Tesla pushes too aggressively and suffers high-profile failures, it could undermine public trust in autonomous vehicles as a whole, potentially setting back the entire industry.

Looking ahead, Tesla aims to scale up rapidly if the initial trial proves successful. Future plans include the introduction of its futuristic-looking purpose-built vehicles, such as the two-door Cybercab and the larger Robovan, both without traditional steering wheels and pedals. Musk’s long-term vision also includes allowing private Tesla owners to put their cars to work as robotaxis—something he has promised would happen “next year” for several years running.

Tesla’s ambitions for the robotaxi business are nothing short of vast, with Elon Musk estimating that autonomous vehicles could add as much as $10 trillion to Tesla’s market capitalisation. But the road ahead will not be easy. Public trust must still be earned, and rivals like Waymo are already making significant headway. Tesla has to address technical glitches, prove its camera-only approach is safe, and compete with established rivals while being watched closely by investors, regulators, and the public alike. Whether Tesla can transform the promise of robotaxis into a global reality remains an open question—one that could define the next era of mobility.

---

Update 30.06 - the text has been updated with a paragraph about Waymo’s issues

If you enjoy this post, please click the ❤️ button or share it.

---

Do you like my work? Consider becoming a paying subscriber to support it

For those who prefer to make a one-off donation, you can 'buy me a coffee' via Ko-fi. Every coffee bought is a generous support towards the work put into this newsletter.

Your support, in any form, is deeply appreciated and goes a long way in keeping this newsletter alive and thriving.

---

🦾 More than a human

A New BCI Instantly Synthesizes Speech
Researchers have developed a brain-computer interface (BCI) capable of synthesising near-instantaneous speech by decoding neural signals in real time, offering new hope for people who have lost the ability to speak due to neurodegenerative diseases like ALS. The system, tested on a patient with ALS, could reproduce not only words but also vocal intonation and emotional inflexion. Though still in early stages, with around 56% intelligibility and limited everyday usability, the technology marks a major step towards restoring naturalistic speech through neural decoding.

🧠 Artificial Intelligence

Mira Murati’s Thinking Machines Lab closes on $2B at $10B valuation
Thinking Machines Lab, a secretive AI startup founded by Mira Murati, former CTO of OpenAI, has reportedly raised $2 billion at a $10 billion valuation, according to the Financial Times. If accurate, this could represent the largest seed round in history.

Apple Executives Have Held Internal Talks About Buying AI Startup Perplexity
Apple is reportedly exploring a potential acquisition or partnership with Perplexity AI, as part of its broader push to strengthen its AI capabilities amid growing competition. Additionally, if successful, a deal with Perplexity could end Apple’s $20 billion search deal with Google, which is now under threat from US antitrust enforcers. While internal discussions remain preliminary and no formal offer has been made, such a move could support Apple’s efforts to build its own AI-based search engine or to incorporate Perplexity technology in Safari and Siri. However, Apple is not alone—Samsung and Meta are also eyeing a partnership with Perplexity or trying to acquire the startup.

Pope Leo Takes On AI as a Potential Threat to Humanity
The newly elected Pope Leo XIV has made AI a central focus of his papacy. He explained that, just as his namesake Leo XIII stood up for the rights of factory workers during the Gilded Age, he now seeks to address the challenges of today’s digital revolution. In his first address, he called for urgent action on the ethical issues posed by AI, urging international regulation to protect human dignity, justice, and labour. Under Leo XIV’s leadership, the Catholic Church aims to play a leading role in shaping global debates on the risks and responsibilities of AI.

Anthropic Scores a Landmark AI Copyright Win—but Will Face Trial Over Piracy Claims
Anthropic has scored a major legal victory after a US federal judge ruled that the AI company’s use of copyrighted materials to train its models qualifies as “fair use.” Judge William Alsup described the training as “transformative,” aligning it with copyright’s purpose of fostering innovation. The verdict could influence numerous ongoing AI copyright cases. However, the court also found that Anthropic had illegally amassed a vast library of pirated books—over seven million—and allowed a trial to proceed over this alleged infringement, potentially exposing the company to billions in damages.

Federal judge sides with Meta in lawsuit over training AI models on copyrighted books
A US federal judge ruled in favour of Meta in a lawsuit brought by 13 authors who alleged the company illegally trained its AI on their copyrighted books. Judge Vince Chhabria found Meta’s actions fell under “fair use,” noting the use was transformative and the plaintiffs failed to show market harm. However, the ruling was limited in scope, with the judge emphasising it does not establish that all AI training on copyrighted material is lawful, and that stronger cases could succeed with better evidence.

Court filings reveal OpenAI and io’s early work on an AI device
If you visit the announcement page for the OpenAI and Jony Ive partnership, you’ll notice it is temporarily unavailable due to a court order resulting from a trademark complaint by iyO, a Google-backed hardware startup developing custom-moulded earpieces. As The Verge reports, IyO alleges OpenAI knew about its technology and shared confidential meetings before launching “io,” prompting the lawsuit after OpenAI declined to invest. OpenAI has since removed the “io” branding and is seeking to dismiss the case, while maintaining its partnership with Ive. Court documents and leaked emails reveal that OpenAI and Ive’s team spent the past year researching in-ear hardware, but their first device will not be an in-ear or wearable product.

What comes after the smartphone?

Masa Son Pitches $1 Trillion US AI Hub to TSMC, Trump Team
SoftBank founder Masayoshi Son has proposed “Project Crystal Land”—a trillion-dollar industrial complex in Arizona, envisioned to emulate China’s manufacturing hub Shenzhen and serve as a centre for robotics and AI. Son hopes to partner with TSMC and other tech giants, but the project’s success hinges on major industry interest and government support.

OpenAI designs rival to Office and Workspace, The Information reports
OpenAI is reportedly developing new productivity tools for ChatGPT, such as collaborative document editing and integrated chat, to rival Google Workspace and Microsoft Office. These tools reflect the company’s ambition to make ChatGPT a comprehensive personal and workplace assistant, and come amid ongoing negotiations with Microsoft, which holds a major stake in OpenAI. No official launch dates have been announced.

The ‘OpenAI Files’ push for oversight in the race to AGI
The OpenAI Files is a new project launched by the Midas Project and the Tech Oversight Project to collect and publish concerns about governance, leadership, and culture at OpenAI as the company races towards developing artificial general intelligence (AGI). The Files highlight issues such as weakened safety processes, potential conflicts of interest involving Sam Altman, and changes to profit structures favouring investors over public benefit. The initiative aims to promote responsible governance and ethical leadership in AI, arguing that as AGI nears, the public deserves transparency and a say in decisions with far-reaching consequences.

Introducing Gemma 3n
Google has fully released Gemma 3n, a new model from its Gemma family of open models. Gemma 3n is a multimodal model, natively supporting image, audio, video, and text inputs and outputs, and is designed with efficiency in mind. It features novel components such as the MatFormer architecture for compute flexibility, Per Layer Embeddings (PLE) for memory efficiency, and new audio and MobileNet-v5-based vision encoders optimised for on-device use cases. According to benchmark results provided by Google, Gemma 3n achieves an LMArena score of over 1300—better than Llama 4 Maverick and OpenAI GPT-4.1-nano—making it the first model under 10 billion parameters to reach this benchmark.

Imagen 4 is now available in the Gemini API and Google AI Studio
Imagen 4, Google’s “best text-to-image model yet”, is now available in paid preview via the Gemini API and for limited free testing in Google AI Studio, with general availability expected in the coming weeks. Alongside Imagen 4, Google has also released Imagen 4 Ultra, which is better at following instructions and is designed to produce outputs that are more closely aligned with text prompts. According to Google, this results in strong performance compared to other leading image generation models.

Saab achieves AI milestone with Gripen E
Saab has announced that it has successfully completed the first three flights of a Gripen E fighter jet equipped with Centaur, Helsing’s AI agent. As part of Saab’s “Project Beyond,” the AI autonomously executed complex combat manoeuvres and supported the pilot in simulated combat scenarios, marking a major step forward for AI integration in military aviation.

▶️ Andrej Karpathy: Software Is Changing (Again) (39:31)

If you are interested in software development, I highly recommend this talk from Andrej Karpathy, where he outlines his predictions for the future of software in the era of AI. Karpathy explains how large language models (LLMs) are fundamentally changing the way software is built, moving beyond traditional code to systems that can be programmed using natural language prompts. He sees a bright future for programming and says that now is an amazing time to enter the software industry.

🤖 Robotics

Gemini Robotics On-Device brings AI to local robotic devices
Google has introduced Gemini Robotics On-Device, a VLA (vision-language-action) model optimised to run efficiently directly on robotic hardware, without relying on cloud-based servers. Gemini Robotics On-Device enables real-time performance in latency-sensitive scenarios and supports a range of robot types, from bi-arm systems to humanoids. According to benchmark results provided by Google, the new model outperforms the previous best on-device model (the model wasn’t named) and approaches the performance of the full Gemini Robotics model.

NEURA launches new cognitive robots and robotics ecosystem
At Automatica 2025, NEURA, a German robotics company, announced a suite of major advancements, including the world premiere of its third-generation 4NE1 humanoid robot, the market launch of the MiPA cognitive household and service robot, and the introduction of the Neuraverse open robotics ecosystem. NEURA aims to make robotics accessible for both industry and everyday life, with ambitions to deliver five million robots by 2030 and become a global leader in the sector.

Nvidia, Foxconn in talks to deploy humanoid robots at Houston AI server making plant
Reuters reports that Foxconn and Nvidia are in advanced talks to deploy humanoid robots at a new Foxconn factory in Houston, where Nvidia’s GB300 AI servers will be produced. The robots—potentially developed by Foxconn or China's UBTech—are expected to handle tasks such as assembly and cable insertion. Deployment is targeted for early next year.

▶️ iRonCub3: First Liftoff of a Jet-Powered Humanoid Robot (1:01)

Well, I did not expect to see a jet-powered humanoid robot, but here we are. Developed by researchers at Istituto Italiano di Tecnologia, iRonCub3 is a humanoid robot with jet engines instead of hands, which allowed the robot to fly 50 cm off the ground, while maintaining dynamic stability, thanks to advanced AI-based control systems and aerodynamic modelling.

Generalist: An early preview of model capabilities

Generalist, a robotics startup founded by former engineers from Google, Boston Dynamics, and OpenAI, has shared an early preview of its model. In the demonstrations, robotic arms—controlled by end-to-end deep neural networks—perform complex, dexterous tasks requiring fine motor manipulation, bimanual coordination, and precise control, showcasing adaptability, robustness, and cross-platform generalisation without task-specific training.

Danish military using robotic sailboats for surveillance in Baltic and North seas
Denmark has launched four uncrewed robotic sailboats, called “Voyagers”, for a three-month trial to boost maritime surveillance in the Baltic and North Seas. These wind and solar-powered vessels will patrol Danish and NATO waters, providing continuous monitoring at a fraction of the cost of traditional warships. The trial comes as NATO seeks to increase security following recent incidents of undersea cable damage and hybrid attacks, with the Voyagers forming part of a new layered surveillance system.

🧬 Biotechnology

AlphaGenome: AI for better understanding the genome
Researchers from DeepMind have unveiled AlphaGenome, a powerful new AI model capable of predicting how genetic variations in human DNA affect gene regulation and biological processes with unprecedented accuracy. By analysing DNA sequences up to one million base pairs long and providing high-resolution, base-level insights, AlphaGenome outperforms previous models and offers a unified tool for understanding genetic variants’ effects, especially in non-coding regions linked to disease. Now available for non-commercial research via API, AlphaGenome is expected to accelerate advances in genomics, disease research, and synthetic biology, though it is not yet suitable for personal genome prediction or clinical use.

FDA approves powerful HIV drug that nearly eliminated spread in clinical trials
The US Food and Drug Administration has approved Yeztugo, an HIV-prevention drug from Gilead Sciences, which has shown near-total effectiveness in clinical trials with twice-yearly injections. Yeztugo (generic name lenacapavir) offers a significant advantage over daily oral PrEP (pre-exposure prophylaxis) by addressing adherence challenges and dramatically lowering HIV transmission rates. While hailed as a transformative step in HIV prevention, access may be hindered by its high cost, insurance hurdles, and recent political efforts to slash HIV-prevention funding.

▶️ Giving your house plants genetically enhanced super powers (26:25)

In this video, The Thought Emporium—YouTube’s chief mad bioengineer—checks how Neoplants use various processes and techniques to create their genetically engineered plants designed to purify air. To be clear, the video is essentially a giant ad for Neoplants, but it is interesting to see how a group of genetic engineers are turning a common houseplant into something more than just a plant.

Thanks for reading. If you enjoyed this post, please click the ❤️ button or share it.

Humanity Redefined sheds light on the bleeding edge of technology and how advancements in AI, robotics, and biotech can usher in abundance, expand humanity's horizons, and redefine what it means to be human.

A big thank you to my paid subscribers, to my Patrons: whmr, Florian, dux, Eric, Preppikoma and Andrew, and to everyone who supports my work on Ko-Fi. Thank you for the support!

My DMs are open to all subscribers. Feel free to drop me a message, share feedback, or just say "hi!"