A chill wind hits AI and the markets - Sync #545
Plus: GPT-5.1; Yann LeCun leaves Meta; Waymo drives on freeways in California; riding in a Chinese robotaxi; tech billionaires invest in designer babies startups; autonomous racing cars; and more!
Hello and welcome to Sync #545!
Our main story this week looks at signs that the AI boom may be losing steam, as tech stocks wobble and SoftBank makes an unexpected move that has investors questioning how long the current momentum can last.
Elsewhere in AI, OpenAI released GPT-5.1, Anthropic announced a $50 billion investment in US AI infrastructure, and Yann LeCun, one of the Godfathers of AI, is leaving Meta. Meanwhile, Fei-Fei Li, the Godmother of AI, explains why spatial intelligence is the next frontier in AI and releases her first AI product.
Over in robotics, Waymo can now use freeways in the US, a home humanoid robot from China, and the Abu Dhabi Autonomous Racing League concluded its second season.
In other news, tech billionaires are investing in startups working on designer babies, Anthropic is on track to be profitable sooner than OpenAI, an AI persona tops the US Billboard charts, a story on how Singapore failed to grow more of its food, and more!
Enjoy!
A chill wind hits AI and the markets
ChatGPT entered the stage almost exactly three years ago. Since then, billions of dollars have been invested in AI companies, and even larger sums are flowing into building data centres and infrastructure to serve AI models at a massive scale. Startups that did not exist two or three years ago have reached multibillion-dollar valuations. OpenAI, the poster child of the generative AI boom, has become the world’s most valuable private company. Other tech companies saw their valuation skyrocket, with Nvidia being the biggest winner in this game by recently becoming the first company to reach $5 trillion valuation.
On the surface, the future looks bright. But cracks have started to show. Investors’ optimism in AI has begun to fade, and concerns about the AI bubble are growing louder. At this point, the question is not if but when the bubble will burst. We don’t know what will trigger it, but developments from the past two weeks suggest that we may be getting close.
Nasdaq’s worst week since April
Recently, the Nasdaq recorded its worst week since President Trump launched a trade war in April. The tech-heavy Nasdaq Composite Index, which had been fuelled by AI enthusiasm, fell 3% in the first week of November. Optimism faded as investors began to worry that big tech companies were overspending and overpromising on their AI initiatives.
The week opened on a hopeful note. Amazon’s $38 billion cloud deal with OpenAI briefly boosted confidence, and early reactions to Amazon’s and Alphabet’s earnings were positive. But momentum quickly reversed. Palantir’s quarterly results underwhelmed Wall Street, raising questions about whether huge AI investments will actually generate meaningful returns. Palantir—despite more than doubling in value earlier this year—fell 11% during the week. Nvidia declined 7% and Oracle lost 9%. While investors initially welcomed results from Amazon and Alphabet, they were disappointed by Meta and Microsoft. By the end of the week, only Amazon managed a small gain. Meta and Microsoft, both signalling plans to continue heavy AI spending, ended the week down roughly 4%.
However, the less tech-heavy S&P 500 and Dow Jones Industrial Average dropped only by 1.6% and 1.2%, respectively, suggesting that the pullback was concentrated among the companies most associated with the tech industry.
These moves took place against a wider backdrop of unease. Consumer sentiment fell to its lowest level since mid-2022. Job-cut announcements nearly tripled in October. The US government shutdown also halted the release of key economic data, removing information that might have reassured investors. Instead, the lack of data only added to uncertainty.
SoftBank goes all in on OpenAI
While Wall Street grew more sceptical, one of the world’s most aggressive tech investors made a bold move. SoftBank, led by Masayoshi Son, disclosed that it had sold its entire $5.8 billion Nvidia stake in October. The timing surprised the market, as SoftBank exited only slightly below Nvidia’s all-time high. Nvidia’s stock dropped 2% following the news.
According to SoftBank, the sale was not a sign of reduced confidence in Nvidia or in AI more broadly. Instead, the company says it needed cash to fund its accelerating investment in OpenAI. SoftBank plans to invest more than $22 billion in OpenAI this quarter alone, with total investment expected to reach nearly $35 billion by year-end. To finance this, it has sold part of its T-Mobile stake for $9.17 billion, issued debt in multiple currencies, and taken out several bridging loans.
Masayoshi Son has a long history of making large, high-stakes bets. Some, like his early investment in Alibaba, paid off immensely. Others, like WeWork, resulted in costly failures. His latest move has split opinion. Some analysts argue that reducing exposure to Nvidia hints at expectations of softer GPU demand in the coming years. Others believe SoftBank simply wants to concentrate capital on OpenAI because it sees the company as the next major growth engine. So far, SoftBank’s bet on OpenAI is paying off. The company reported that its second-quarter net profit more than doubled to $16.6 billion, mainly driven by its stake in OpenAI.
Either way, SoftBank’s actions have added uncertainty to an already fragile market. The company’s share price has swung sharply in recent days, falling as much as 10% after the announcement. The move has only intensified existing concerns about a potential AI bubble.
Warning shots
These events are not the first warning signs. Earlier this year, DeepSeek released R1, an open AI model that approached the performance of top proprietary models from OpenAI, Anthropic and Google while being more efficient. The release of R1 briefly wiped $600 billion off Nvidia’s valuation. In August 2025, a report from MIT found that 95% of generative AI projects have failed. The report itself is more nuanced, but the headline made waves and further shook investors’ faith in AI.
The AI bubble hasn’t burst yet. But we don’t know what will cause it to burst or when it will happen. It could be weeks or months away. We might be just a paper or two away from some Chinese AI inventing a more efficient architecture that calls into question the current US emphasis on scaling infrastructure. Or it could be something else—a disappointing earnings report, a policy surprise, or a sudden shift in sentiment—that sparks panic on Wall Street and triggers a large-scale sell-off. Markets are irrational, after all.
What we do know is that the warning signs are becoming harder to ignore. The momentum that carried AI valuations to sky-high levels is fading, and the industry may be nearing the point where expectations and reality collide. A correction may not be imminent, but it is starting to look increasingly likely.
If you enjoy this post, please click the ❤️ button and share it.
🦾 More than a human
Genetically Engineered Babies Are Banned. Tech Titans Are Trying to Make One Anyway.
Silicon Valley investors and billionaires, including Sam Altman and Brian Armstrong, are funding startups that aim to screen or even edit human embryos to reduce disease risk or influence traits such as intelligence—something banned in the US and criticised by many scientists as unsafe. Supporters say these technologies could lead to healthier children, while critics warn they are unproven, risky, and could open the door to modern eugenics.
The Complicated Reality of 3D Printed Prosthetics
This article explores why 3D printing has not made prosthetic limbs as cheap or accessible as many once hoped. Early excitement overstated what the technology could achieve, and although today’s 3D-printed prosthetics can be precise and well-fitting, they remain expensive to produce. Efforts to create low-cost prosthetics are often blocked by insurance rules and market pressures, so the main obstacles lie in the wider prosthetics ecosystem rather than in the technology itself.
🧠 Artificial Intelligence
GPT-5.1: A smarter, more conversational ChatGPT
OpenAI has released GPT-5.1, which the company says offers improved intelligence, usability, customisation, and developer tooling. The new model comes in two variants: Instant, which is more conversational and responsive, and Thinking, which adapts its reasoning time to task complexity—alongside improved personalisation with new tone presets and real-time style controls. The update rolls out first to paid users. For developers, GPT-5.1 adds adaptive reasoning in the API, a no-reasoning mode for low-latency tasks, 24-hour prompt caching, improved coding performance, and two new tools: apply_patch for structured code edits and shell, which allows the model to interact with a local computer through a controlled command-line interface. GPT-5.1 is available to developers on all paid tiers in the API, with the same pricing and limits as GPT-5.
Anthropic Is on Track to Turn a Profit Much Faster Than OpenAI
This article compares the financial paths of Anthropic and OpenAI, showing that Anthropic is likely to become profitable sooner because it keeps its costs closer to its revenue and focuses mainly on selling to businesses. OpenAI, meanwhile, is spending far more on chips, data centres and new products, leading to much larger losses and higher cash burn as it pursues rapid growth. Despite similar early cash-burn rates relative to revenue, Anthropic’s costs are projected to fall sharply, enabling it to break even by 2028, whereas OpenAI expects rising losses and heavy cash burn until at least 2030.
Anthropic invests $50 billion in American AI infrastructure
Anthropic is investing $50 billion to build new, highly efficient data centres with Fluidstack in Texas and New York, adding thousands of new jobs while contributing to the Trump administration’s AI Action Plan. The facilities, due to open in 2026, will provide the infrastructure needed for Anthropic’s frontier research and to meet fast-growing demand for Claude.
Meta’s chief AI scientist Yann LeCun reportedly plans to leave to build his own startup
According to the Financial Times, Meta’s chief AI scientist, Yann LeCun, one of the Godfathers of AI, plans to leave the company to start his own startup focused on world-model AI systems. His departure comes as Meta reorganises its AI efforts through major hires and investments, moves that have caused internal tension and shifted attention away from the long-term foundational research LeCun has led at Facebook Artificial Intelligence Research (FAIR).
From Words to Worlds: Spatial Intelligence is AI’s Next Frontier
Fei-Fei Li, an AI pioneer who is often referred to as the Godmother of AI, makes a case that the next major frontier in AI is spatial intelligence—the ability for AI to understand and reason about the physical world. She argues that spatial intelligence is the missing pillar for truly intelligent machines. With it, AI could profoundly improve science, creativity, robotics, and human wellbeing, bringing us closer to the original vision of AI.
Fei-Fei Li’s World Labs speeds up the world model race with Marble, its first commercial product
World Labs, an AI startup founded by Fei-Fei Li, has released Marble, a generative AI tool that turns text, images, and videos into editable, downloadable 3D worlds. Unlike other models that change as you move through them, Marble builds stable scenes that users can edit, expand, and export with new tools like a simple 3D layout editor.
OpenAI, Apple Lose Bid to Toss Musk xAI Suit Over Competition
A US judge has allowed Elon Musk’s X Corp. and xAI to move forward with a lawsuit claiming Apple and OpenAI worked together to limit competition in the AI market by integrating OpenAI tools into the iPhone. The companies deny the accusations, saying Apple isn’t tied to OpenAI exclusively and suggesting Musk’s claims stem from his ongoing feud with OpenAI’s leaders.
Murati’s Thinking Machines in Funding Talks at $50 Billion Value
Thinking Machines Lab, founded by former OpenAI executive Mira Murati, is reportedly in early talks to raise new funding that could value the company at around $50–60 billion, far above its $12 billion valuation in July. Launched this year, the startup has hired many former OpenAI staff and released its first product, Tinker, which helps users customise large language models. The potential deal would make the company one of the most valuable private startups.
AI startup Cursor raises $2.3 billion funding round at $29.3 billion valuation
Cursor, an AI startup that makes a popular coding tool, has raised $2.3 billion in a Series D funding round at a $29.3 billion valuation. Additionally, the company announced that it crossed over $1 billion in annualised revenue. Backed by major investors like Accel and Google, Cursor says it will use the new funding to improve its technology and expand the team, while continuing to compete with other big AI coding tools from companies such as OpenAI and Anthropic.
Breaking Rust: AI artist tops US chart for first time as study reveals alarming recognition stats
Breaking Rust, an AI-generated country persona, has topped a US Billboard chart, sparking renewed concerns about the growing presence of AI in music as many listeners fail to realise the “artist” isn’t human. This coincides with a Deezer–Ipsos study that shows that 97% of people can’t tell AI-made music from human-made tracks, raising worries about how AI could harm real musicians, weaken copyright protection and lead to more generic-sounding music.
OpenAI used song lyrics in violation of copyright laws, German court says
A Munich court ruled that OpenAI’s ChatGPT broke German copyright law by reproducing lyrics from nine protected songs. The case could set an important precedent in Europe for how AI companies use copyrighted material. The court ordered OpenAI to pay damages, though the amount was not disclosed. OpenAI disagrees with the decision and may appeal.
SIMA 2: An Agent that Plays, Reasons, and Learns With You in Virtual 3D Worlds
SIMA 2 is DeepMind’s new AI agent that can play and learn in 3D video games. It expands on the original SIMA by using Gemini to reason, and it understands high-level goals, follows instructions, chats naturally and learns both from people and from exploring games on its own. It works across many different game worlds, not just the ones it was trained on. DeepMind says SIMA 2 is a step toward general embodied intelligence—agents that not only think but act in complex 3D (and potentially physical) worlds—though it still struggles with very long, complex tasks. SIMA 2 is currently a research preview, available to a small cohort of academics and game developers.
Tesla AI boss tells staff 2026 will be the ‘hardest year’ of their lives in all-hands meeting
Business Insider reports that Tesla’s AI chief, Ashok Elluswamy, told staff that 2026 will be an extremely challenging year as the company pushes to expand its Robotaxi service and begin production of the Optimus humanoid robot. The next year is seen as a critical test for the company. Additionally, Elon Musk’s recently approved $1 trillion pay package depends heavily on achieving milestones in both programmes. According to the report, the meeting was framed as a rallying call, urging workers to prepare for intense workloads as Musk remains closely involved with both projects.
▶️ The Chinese AI Iceberg (27:06)
The Chinese AI ecosystem is booming. Open source AI models are mostly dominated by Chinese models, which are getting closer to the performance of proprietary models from the likes of OpenAI, Anthropic or Google. This video breaks down the current Chinese AI landscape—from flagship labs to underground research groups—using the Iceberg Format, the internet’s favourite method for explaining topics from the well-known to the obscure.
▶️ Anthropic: We taught robot dogs to fetch (7:39)
To show how good Claude is at coding, researchers at Anthropic asked their colleagues, who had never programmed a robot, to program one. The participants were divided into two groups—one team was using Claude, while the other was not. Both teams had to make the robot fetch a ball—first manually, then with code, and finally by programming the robot to fetch the ball autonomously. The result was that the team using Claude moved faster, as it helped them write code, fix errors, and set things up, while the other team spent more time struggling with technical issues.
🤖 Robotics
Waymo robotaxis are now giving rides on freeways in LA, San Francisco, and Phoenix
Waymo is starting to offer robotaxi rides that use freeways in San Francisco, Phoenix, and Los Angeles, which could cut travel times by up to half and enable airport trips and longer journeys. Initially available only to users who opt in via the app, the new capability helps create a unified 260-mile service area across the Peninsula, allowing Waymo to operate from San Francisco down to San Jose. Waymo says it carried out extensive testing to make sure its vehicles can safely handle motorway driving and work smoothly with local authorities.
Riding in a Chinese Robotaxi Is Pretty Smooth—That’s a Problem for Waymo
This article takes us to the streets of Beijing to experience Chinese robotaxis. Companies like Baidu, Pony AI and WeRide are quickly growing their robotaxi services, offering smooth, fully driverless rides at lower costs than US rivals. They are also expanding into Europe, Asia and the Middle East through partnerships with Uber and Lyft, though they face competition and safety concerns. Despite the rapid progress, the businesses are still unprofitable, and Chinese robotaxis are unlikely to enter the US because of tariffs and data-security issues.
A2RL Season 2 Finals Show A Lot Can Happen In A Short Time
The second season of the Abu Dhabi Autonomous Racing League (A2RL) ended this weekend with a world-first six-car autonomous final at Yas Marina Circuit. TGM won the time-trial event, and the Human vs AI challenge was far closer than last year, with Daniil Kvyat, a former F1 driver, finishing less than a second behind TUM’s AI car. In the main race, Unimore briefly led before a collision ended its chances, allowing TUM to take the 2025 title. Overall, the event proved how much progress has been made in just 18 months and set strong expectations for the 2026 season.
▶️ MindOn turned Unitree G1 into a robotic housekeeper (2:22)
Two weeks ago, 1X launched pre-orders for NEO, its humanoid robot for the home. NEO gathered mixed responses at best, especially after it was revealed that most of what it has been shown doing was not autonomous (you can read more about 1X NEO here). This week, Chinese company MindOn showed its own home humanoid robot. Based on Unitree G1, the robot is shown carrying out various household tasks, such as watering plants, cleaning, and picking up packages. MindOn claims the footage was not sped up and that the robot was not being teleoperated.
▶️ Toward Generalist Humanoid Robots (1:17:56)
In this lecture, Yuke Zhu explains how recent AI advances and new ways of using data are bringing us closer to generalist humanoid robots that can handle everyday tasks. He describes how combining real robot data, simulation, and human videos allows foundation models to learn useful and transferable skills, supported by new tools for generating synthetic training data and methods for aligning it with real-world behaviour. While challenges remain, such as dexterity, whole-body control, and safety, Zhu argues that progress across AI, hardware, and data is accelerating—and that humanoids are not a question of if, but when.
🧬 Biotechnology
Colossal Biosciences Acquires Viagen, the Leader in Animal Cloning
Colossal Biosciences, a biotech company known for its efforts to bring back extinct species, has acquired Viagen Pets and Equine, a leading global animal cloning company. It is Colossal’s first acquisition since launching in 2021. Viagen will continue operating as usual but will use Colossal’s resources to expand its cloning and cryopreservation work, especially for endangered animals. The terms of the acquisition have not been disclosed.
Lilly Taps Insilico Medicine’s AI in $100M Drug Discovery Collaboration
Insilico Medicine has expanded its partnership with Eli Lilly in a new research and licensing deal worth over $100 million that uses Insilico’s Pharma.AI platform to design and optimise new drugs. The collaboration builds on a 2023 software deal and draws on Insilico’s fast-growing pipeline, which includes 22 preclinical candidates and several advancing clinical programmes.
A Gene Editing Therapy Cut Cholesterol Levels by Half
A small early-stage trial by Crispr Therapeutics found that a one-time CRISPR treatment designed to switch off the ANGPTL3 gene cut LDL cholesterol and triglycerides by about 50% in participants receiving the highest dose, with effects lasting at least 60 days and only minor side effects reported. Although early, the results suggest that gene editing could one day offer long-lasting treatment for common heart disease risks.
Gene Therapy Aims To Slow Huntington’s Disease To A Crawl
This article reports on a promising new genetic therapy for Huntington’s disease, a hereditary condition that leads to brain degeneration. The treatment uses a specially engineered virus inserted into key areas of the brain, which helps cells produce microRNA that reduces the harmful mutant huntingtin protein. Early trial results in 29 patients show the therapy could slow disease progression by up to 75%, preserve brain tissue, and significantly improve quality of life, although the procedure is highly invasive and costly.
💡Tangents
▶️ Singapore Tried to Grow More of Its Own Food... (24:09)
This video tells the story of when Singapore tried to grow 30% of its nutritional needs locally by 2030 without greatly expanding farmland. The story includes Singapore’s recent agricultural policies, the attempt to recreate its success in securing its water supply but in food production, high-tech indoor vertical farms, deep-sea aquaponics, lab-grown meat, and why the “30-by-30” project eventually failed to deliver on its goals.
Thanks for reading. If you enjoyed this post, please click the ❤️ button and 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!"