Meta Goes All In on Personal Superintelligence - Sync #528
Plus: Cognition acquires Windsurf; OpenAI releases ChatGPT Agent; Figure teases a new humanoid robot; US DoD awards contracts to AI companies; lab-grown salmon gets FDA approval; and more!
Hello and welcome to Sync #528!
This week, we take a closer look at Meta’s ambition to build personal superintelligence, its push to quickly build massive AI data centres and the possibility that Meta may abandon its open-source approach to AI.
Elsewhere in AI, Cognition—the maker of AI coding agent Devin—has acquired what remains of Windsurf. Meanwhile, OpenAI has announced ChatGPT Agent, Amazon has released a suite of tools and services for building AI agents, and the US Department of Defense has awarded contracts worth up to $200 million to Anthropic, Google, OpenAI, and xAI. Speaking of xAI, the Grok app now features AI companions, one of which is an anime goth girl. Additionally, a former OpenAI employee reflects on their time at the company, rumours have begun to emerge about Anthropic’s next funding round, and a story on how Builder.ai became the biggest scam in AI (so far).
Over in robotics, Figure has teased its F.03 humanoid robot. We also have a young startup building an open-source humanoid robot, and an army of robots moving an entire city block in Shanghai.
This week’s issue of Sync also features lab-grown salmon meat being approved by the FDA, a trip into Neuralink’s competitor’s secretive labs, and the news of eight healthy babies born after IVF using DNA from three people.
Enjoy!
Meta Goes All In on Personal Superintelligence
Having spent mountains of money poaching top AI talent from rivals, Meta is now taking its next big steps in the quest to build “personal superintelligence.”
Unlike OpenAI and Google, who are racing to automate business workflows and replace knowledge workers, Meta’s big bet is something it calls “personal superintelligence.”
Rather than aiming AI at the enterprise market, Meta wants to create a true digital companion—something that helps with your relationships, memory, creativity, and day-to-day life. Imagine an AI that sees what you see, hears what you hear, and quietly manages the things you care about most. Imagine smart glasses that remember every conversation, or an AI that follows up on everything you mention but forget to act on. That’s the sort of personal superintelligence Meta is chasing.
Here’s how Zuckerberg put it in a recent interview:
What are the things that people care about in their own lives? Some of that is productivity. But a lot of it is relationships and culture and creativity and having fun and enjoying life.
To make that vision a reality, Meta needs three things: world-class talent, massive computing power, and enormous amounts of data. Data has never been a problem for Meta, thanks to Facebook, Instagram, and WhatsApp. Over the past year, the company has been assembling a dream team of AI researchers and engineers, luring talent from OpenAI, Apple, Anthropic, and Google with multi-million dollar packages (I wrote more about this in a previous issue). That leaves Meta with the third pillar for a successful AI project—access to computing power.
Meta Builds Its Titans: Hyperion and Prometheus
Meta’s competitors are already ahead in computing power. Google and Microsoft have vast data centre networks loaded with top-of-the-line Nvidia GPUs and, in Google’s case, custom AI chips. OpenAI is building its first Stargate AI data centre in Texas, xAI has Colossus—already at 200,000 GPUs and aiming for a million—and Amazon is building a new AI data centre with Anthropic in mind.
To compete, Meta is building two of the most ambitious data centres in the world—Hyperion and Prometheus. Hyperion alone, according to Zuckerberg, will cover a footprint comparable to most of Manhattan and deliver up to five gigawatts of computing power. The first two gigawatts are planned to come online by 2030 at a site in Richland Parish, Louisiana, with further expansion over the following years. Prometheus, a one-gigawatt super cluster located in New Albany, Ohio, is set to become operational in 2026, making Meta one of the first tech companies to have an AI data centre of this size.
Additionally, Meta is thinking outside the box and using non-traditional, hurricane-proof tent structures to accelerate the buildout of these data centres. This sense of urgency reflects the breakneck pace of the AI arms race, where every month could be the difference between leading or falling irreversibly behind.
Will Meta Abandon Open Source?
One of the most consequential decisions facing Meta is whether to continue its tradition of open-sourcing its most advanced AI models. Historically, Meta’s open models—such as the Llama family—fuelled a boom in AI research and experimentation around the world. Open source has become a kind of calling card for Meta, setting it apart from rivals like OpenAI and Anthropic, who keep their best models closed.
Behind the scenes, however, Meta is quietly debating whether to shift away from its open-source approach to AI. According to a report from The New York Times, Alexandr Wang, the lab’s chief, has floated the idea of abandoning open source for a closed approach, mirroring the business models of its competitors like OpenAI or Anthropic. If this were to happen, it would be not only a massive philosophical shift but also a strategic one.
From a business perspective, giving away your best AI models for free isn’t exactly a good strategy. Training these models costs tens or even hundreds of millions of dollars. That’s why companies like OpenAI and Anthropic keep their top models closed—being the gatekeeper is central to their business. Meta, on the other hand, has been able to release models like Llama for free because its business relies on advertising, not on charging for AI access. But as AI becomes a bigger part of Meta’s core business—and with the push towards personal superintelligence—shifting to a closed-source approach might make more sense for Meta going forward.
If Meta abandons open source, who will fill the void? Google and Microsoft both offer smaller open models (Gemma and Phi), but their top-tier models remain closed. Mistral, a French startup, has made its name by releasing open models, and OpenAI keeps promising—but delaying—its first open model in years.
However, if Meta pulls back from open source, it could be the perfect opening for Chinese AI companies to step in—and they’re already moving fast. Earlier this year, DeepSeek R1 shook up the open-source AI landscape, proving that Chinese companies aren’t far behind their US counterparts. Just last week, another Chinese startup, Moonshot AI, released Kimi K2, an open-source model that’s not only cheaper but has outperformed OpenAI’s GPT-4.1 and Anthropic’s Claude Opus 4 on some benchmarks.
While American AI companies are getting more secretive—locking down their most powerful models—Chinese companies are heading in the opposite direction, open-sourcing not just small models, but their very best. (To be fair, not every Chinese company is fully open—Baidu, for example, keeps its top models closed). As I mentioned earlier, this might not make the most sense from a business perspective, but it’s a brilliant way to build soft power and win over developers worldwide.
Meta has played this game too—releasing its Llama models for free and earning a ton of goodwill from researchers. For a while, Meta was seen as the unlikely “hero of open source” among big tech. But if Meta abandons open-source AI, that title will almost certainly go to a Chinese company. And that shift could have real geopolitical implications.
Will the $65 Billion Gamble on Personal Superintelligence Pay Off?
Meta is betting big on personal superintelligence, planning to spend up to £72 billion on AI in 2025 alone—all from its own cash flow. That buys top talent, state-of-the-art data centres, and the freedom to iterate even if the first attempts fail.
If Zuckerberg wants to go all-in, then there’s no one who can stop him. Unlike other big tech CEOs, Zuckerberg holds a special class of shares that gives him outsized voting power, meaning he can push through bold, risky ideas even if Wall Street or Meta’s own board is nervous.
But will Meta’s vision for personal superintelligence take off? Zuckerberg argues it will transform how we connect, create, and enjoy life, not just automate our busywork. The big question is whether people will actually want it, and if they do, whether they’ll want it from Meta, a company known for aggressively monetising user data, or would users rather choose to use OpenAI or Google? (Not that those options are necessarily less problematic.)
Zuckerberg nailed the mobile transition and stumbled on VR and the Metaverse. Now he’s all in on personal superintelligence. Will this gamble pay off, or will it join the list of tech’s most expensive experiments? Only time will tell.
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
Inside the Secret HQ of One of Neuralink's Biggest Competitors
This article takes us to the secretive New York headquarters of Synchron, a brain-computer interface (BCI) startup developing the Stentrode—a unique neural implant inserted via a blood vessel, making it less invasive and potentially safer than competitors. The article explores Synchron’s innovative technology, its rivalry with companies like Neuralink, support from major investors, and integration with platforms such as Apple and Nvidia, while also addressing the ethical, privacy, and societal questions posed by the potential mainstream adoption of BCI technology.
Eight healthy babies born after IVF using DNA from three people
Doctors in the UK have announced the birth of eight healthy babies using a groundbreaking IVF technique that combines DNA from three people to prevent serious inherited mitochondrial diseases. The method involves using genetic material from the mother and father, transferred into a donor egg with healthy mitochondria, to reduce the risk of passing on life-threatening conditions. All eight babies are doing well, offering new hope to families affected by these genetic disorders.
🧠 Artificial Intelligence
Cognition’s acquisition of Windsurf
Cognition, the company behind the coding AI agent Devin, has acquired what remained of Windsurf after Google poached its CEO, co-founder and top talent for $2.4 billion last week. According to Cognition, the acquisition includes Windsurf’s IP, product, trademark and brand, as well as the remaining employees. Cognition did not reveal how much it paid to acquire Windsurf.
Introducing ChatGPT agent: bridging research and action
OpenAI has launched a new agent mode for ChatGPT, allowing the AI to ChatGPT’s new agent mode allows it to autonomously handle complex tasks from start to finish using its own virtual computer. The agent is capable of using a web browser, running code, and analysing data. Benchmark results provided by OpenAI show that the new ChatGPT agent surpasses previous models and human baselines on a range of practical tasks. Interestingly, OpenAI only compared the ChatGPT agent to its own models and omitted comparisons to competitors. The new agent mode is rolling out now to Pro, Plus, and Team users, with wider availability planned in the coming weeks (not currently available in the EU or Switzerland). OpenAI notes that the agent is still in its early stages and can make mistakes, and some features (such as slideshow creation) remain in beta.
Anthropic, Google, OpenAI and xAI granted up to $200 million for AI work from Defense Department
The US Department of Defense has awarded contracts worth up to $200 million to Anthropic, Google, OpenAI, and xAI to accelerate the development and adoption of advanced AI capabilities for national security, as well as address critical national security challenges. “The adoption of AI is transforming the Department’s ability to support our warfighters and maintain strategic advantage over our adversaries,” Doug Matty, the DoD’s chief digital and AI officer, said in a press release.
Thinking Machines Lab Raises a Record $2 Billion, Announces Cofounders
Thinking Machines Lab, an AI startup founded by former OpenAI chief technology officer Mira Murati and several ex-OpenAI employees, has announced a $2 billion seed round, valuing the company at $12 billion. The company is developing advanced multimodal AI to “empower humanity through advancing collaborative general intelligence”, as Murati phrased it, and plans to launch its first open-source product for researchers and startups in the coming months.
SpaceX to Invest $2 Billion Into Elon Musk’s xAI
The Wall Street Journal reports that SpaceX has agreed to invest $2 billion in xAI, accounting for nearly half of xAI’s recent $5 billion equity raise and making this one of SpaceX’s largest-ever investments in another company. This move is part of a reportedly ongoing funding round for xAI, with reports emerging that Elon Musk is seeking new investors in a deal that could value xAI between $170 billion and $200 billion. Musk has been leveraging his business empire to support xAI, recently merging the AI firm with X to further accelerate its growth, as xAI aims to compete with OpenAI, Google, and others in the AI race.
Elon Musk’s Grok is making AI companions, including a goth anime girl
Super Grok users now have access to two AI companions in the Grok app: Ani, an anime goth girl, and Bad Rudy, a fox-like character. As Grok was designed to have a more edgy personality, some users who have tried the new companions have shared some interesting interactions with them.
Amazon considers another multibillion-dollar investment in Anthropic
Amazon is reportedly considering a new multibillion-dollar investment in Anthropic, according to the Financial Times. Amazon has already invested $8 billion into Anthropic, and this move could deepen their strategic partnership and strengthen Amazon’s position as a leading investor ahead of Google. Additionally, reports emerged that Anthropic might be preparing a new funding round, which could value the company at $100 billion.
Introducing Amazon Bedrock AgentCore: Securely deploy and operate AI agents at any scale
Amazon has announced the preview launch of Amazon Bedrock AgentCore, a new suite of tools and services designed to make it easier for businesses to deploy, manage, and scale AI agents. AgentCore offers a range of integrated tools—including secure serverless runtime environments, advanced memory management, observability dashboards, and robust identity controls—that work with any AI framework or model. The platform enables developers to quickly build complex agents for tasks like customer support, workflow automation, and more, while maintaining compliance and operational visibility. Available now in preview in select AWS regions and free to use until 16 September 2025.
Microsoft and OpenAI's AGI Fight Is Bigger Than a Contract
The agreement between Microsoft and OpenAI contains a confirmed contractual clause—also known as “The Clause”—which states that if OpenAI’s AI models achieve artificial general intelligence (AGI), defined as a system that outperforms humans at most economically valuable tasks and is projected to generate profits exceeding $100 billion, OpenAI can deny Microsoft access to its latest models, effectively ending their arrangement and preventing Microsoft from independently developing AGI during the contract’s duration. As the article explains, this clause has become a major source of tension between the two companies, especially as the possibility of AGI appears closer than previously thought. With OpenAI now seeking to restructure its governance and profit model, “The Clause” is at the heart of ongoing renegotiations.
Nvidia is set to resume China chip sales after months of regulatory whiplash
Nvidia has announced plans to resume sales of its H20 AI chips to China after a brief period of US government-imposed export restrictions that were later paused after high-level lobbying and Nvidia’s pledge to invest in US data centres. Apart from H20, Nvidia is also launching a new “RTX Pro” chip designed specifically for the Chinese market.
Apple Will Seriously Consider Buying Mistral
Apple’s venture into AI is not going well—Apple Intelligence is delayed and does not meet expectations, and key people are leaving the company. This article explores an idea shared by Mark Gurman in his Bloomberg newsletter: that Apple may consider acquiring Mistral, a French AI start-up. This could help Apple build its internal AI team, but the article omits one major obstacle: Mistral is French, and France takes digital sovereignty seriously, which can massively complicate a potential acquisition.
The End of Moore's Law for AI? Gemini Flash Offers a Warning
This article analyses Google’s recent move that went mostly unnoticed, but which could signal changes for the AI industry. Google has sharply raised the price of its Gemini 2.5 Flash model, breaking the trend of steadily falling AI costs. The article argues that this price hike reveals the industry has reached a practical cost “soft floor” due to hardware, energy, and demand limitations. As a result, developers should stop expecting AI costs to keep dropping and instead explore more cost-effective options such as batch processing or open-source models.
Reflections on OpenAI
In this post, Calvin French-Owen shares his experience as a software engineer at OpenAI. However, don’t expect to find trade secrets or any spicy stories here. Instead, Calvin tells a story of someone witnessing from inside a company growing extremely fast while under the close eye of the public. He describes OpenAI’s meritocratic, fast-moving environment, where teams operate with a strong bias to action, leadership is visible and involved, and new ideas can emerge from anywhere. Overall, he portrays OpenAI as a uniquely ambitious and mission-driven organisation at the forefront of AI development.
Cloudflare: Introducing pay per crawl: enabling content owners to charge AI crawlers for access
Cloudflare has introduced "pay per crawl," a new system in private beta that allows publishers to charge AI crawlers for access to their online content, rather than simply blocking them or allowing free access. This approach aims to create a fairer, more flexible model for content creators, while also providing them with granular control over which crawlers can access their material and at what price, thereby enabling the direct monetisation of content requests at an Internet scale. Additionally, it addresses technical challenges related to crawler authentication and payment.
Scientists reportedly hiding AI text prompts in academic papers to receive positive peer reviews
Academics from institutions across eight countries have reportedly been hiding prompts for automated peer-review tools to give their preprint papers on platforms like arXiv positive reviews. These prompts, often invisible to human reviewers (for example, by being written in white text on a white background), instruct AI systems not to mention negatives and to provide glowing assessments. This practice highlights concerns about the integrity of scientific publishing amid increasing reliance on AI tools for peer review, but it also illustrates the extent to which academic success is measured by publication output, so much so that some researchers resort to cheating the system.
▶️ Builder.ai - The Greatest AI Scam in History (26:14)
This video documents the rise and fall of Builder.ai, once hailed as a $1.5 billion tech unicorn promising AI-driven app development. The company collapsed after it was revealed that its claims of advanced automation were largely fabricated, with most of the actual work outsourced to low-paid human engineers. Despite attracting huge investments from the likes of Microsoft and SoftBank, and boasting high-profile clients, Builder.ai’s reliance on inflated revenues, deceptive marketing, and questionable business practices was eventually exposed by whistleblowers and journalists, ultimately leading to its dramatic downfall.
🤖 Robotics
▶️ F.03 Battery Development (1:01)
Figure teases the upcoming reveal of its F.03 humanoid robot by showing how it makes the batteries powering the new robot at its BotQ factory.
▶️ Your First Personal Robot (3:50)
Meet K-Scale Labs, a young start-up on a mission to prove that you don’t need an enormous amount of money, nor do you need to be a big tech company, to build a functional and useful humanoid robot. The company has just released its first product, K-Bot, for $10,999. Additionally, K-Scale Labs embraced open-source, which is always nice to see.
▶️ Hundreds of robots move Shanghai city block (1:44)
A small army of robotic feet moved an entire city block in Shanghai. The movement was slow, creeping along at only about 10 metres (33 feet) per day.
🧬 Biotechnology
CRISPR breakthrough allows scientists to edit multiple genes simultaneously
Scientists at ETH Zurich have developed a new CRISPR gene-editing method that can modify dozens of genes simultaneously, whereas previous techniques could typically alter only one gene at a time. Published in Nature Methods, their research shows the new approach—using the Cas12a enzyme and specially designed plasmids—successfully edited 25 gene targets at once and could potentially scale to hundreds, opening the door to large-scale cell reprogramming and the study of complex genetic interactions.
Lab-Grown Salmon Is Here, and It’s Got FDA Approval
Wildtype’s cell-cultivated salmon has become the first lab-grown fish to receive FDA approval in the United States, marking a significant milestone for cell-cultivated seafood. Wildtype’s salmon, praised for its similarity to conventional salmon, will be served at Kann, Portland’s acclaimed Haitian restaurant, and is expected to be introduced at additional restaurants soon.
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!"