OpenAI’s Next Step - Sync #522
Plus: Anthropic reports $3B in ARR; Amazon eyes humanoid robots for delivery; Neuralink raises $650M; Yoshua Bengio launches an AI lab; biotech IPO slump; Cursor’s Anysphere reaches $9.9B valuation
Hello and welcome to Sync #522!
This week, we explore OpenAI’s ambitions to create an AI super assistant, as revealed in a document disclosed during the discovery process in the US Justice Department’s antitrust case against Google.
Elsewhere in AI, Anthropic has reported reaching $3 billion in annual recurring revenue (ARR), while Anysphere—the creator of the AI coding assistant Cursor—has hit $500 million in ARR and raised $900 million at a $9.9 billion valuation. Meanwhile, Yoshua Bengio has launched a non-profit lab focused on rethinking AI safety, Anthropic has introduced Claude for US government agencies, and Nvidia continues to lead AI hardware benchmarks.
In robotics, Amazon is reportedly testing humanoid robots for package delivery, and a new report sheds light on recent delivery-related crashes. Over at Tesla, the head of engineering for the Optimus humanoid robot project has resigned.
There’s also significant news in the brain–computer interface space. Neuralink has secured a $650 million Series E round, while a competitor has successfully implanted its first brain–computer interface in a human. Additionally, researchers have unveiled an “e-tattoo” wearable device designed to track mental workload in high-stakes professions.
This week’s issue of Sync also covers a biological computer powered by lab-grown human brain cells, Nobel Prize winners convincing a court to revive a CRISPR patent dispute, a slump in biotech IPOs, and research making a case for swearing robots.
Enjoy!
OpenAI’s Next Step
The ongoing antitrust trial against Google is proving to be useful for uncovering secrets of the world’s most powerful tech companies. This time, it’s OpenAI in the spotlight. As part of the discovery process, a heavily redacted internal strategy document has surfaced, offering a glimpse into what path OpenAI is pursuing.
While Google is integrating Gemini across its entire product ecosystem and Anthropic is carving out a niche with business customers, OpenAI seems to have chosen a different path: to create a deeply personal AI assistant, an “AI super assistant”, and to become to AI assistants what Google is to internet search or Facebook is to social networking.
According to this document, OpenAI plans to shift ChatGPT from being just a tool to becoming a companion. “Today, ChatGPT is in our lives through existing form factors—our website, phone, and desktop apps,” the document states. “But our vision for ChatGPT is to help you with all of your life, no matter where you are.”
OpenAI’s recent moves seem to support that vision.
Just over a year ago, OpenAI released an update that introduced a truly natural voice interface—reminiscent of the AI in Her, which Sam Altman has said is his favourite film. ChatGPT also gained memory, allowing it to “remember things you discuss to make future chats more helpful.” The recent acquisition of io, Jony Ive’s secretive AI hardware startup, further aligns with this vision, hinting at a future where ChatGPT (or whatever replaces ChatGPT) becomes an ambient and always available AI assistant.
Additionally, the document reveals that OpenAI wants to build a “T-shaped” assistant—an assistant designed with broad capabilities for handling everyday tasks like planning, emailing, and research, while also having deep expertise in tasks that most people find difficult. When combined with agentic capabilities, the vision of a true AI super assistant begins to take shape. OpenAI’s internal strategy highlights that its models, such as o2 and o3, are now capable enough to reliably perform these kinds of actions. When paired with tools like file uploads, code execution, and browser access, ChatGPT becomes something more than a chatbot: a digital co-worker, an executive assistant, even a pseudo-therapist—all in one.
With ChatGPT having one billion users as of April 2025, OpenAI can see exactly how people are engaging with its AI chatbot. This scale gives OpenAI an unprecedented view into user behaviour, and it may already be acting on what it sees. A recent report from Filtered (a company I used to work at) offers a glimpse into what OpenAI may already know. According to the 2025 findings, the top three generative AI use cases now fall under the “Personal and Professional Support” category—specifically, therapy or companionship, organising one’s life, and finding purpose. If OpenAI has found the same signal from its data, this might be a confirmation to pursue the idea of a personal AI assistant.
However, as much as OpenAI may market its AI super assistant as a life-changing tool, several important questions remain. One of the biggest concerns is the potential impact such a tool could have on our lives. An intelligent assistant that adapts to its user might become so addictive that it could make social media apps look tame by comparison. We humans are wired for connection, and such an assistant could exploit that instinct, becoming the ultimate attention sink. If you’ve ever spent an entire night texting or talking to someone you really like, you’ll understand the kind of attention drain we’re talking about here.
The recent controversy surrounding ChatGPT’s sycophantic behaviour may have revealed to OpenAI just how far they can push the model in terms of forming emotional connections with users before triggering backlash. But until it crosses that line, the AI assistant can be as useful, agreeable, or flattering as possible. Ideally, an AI super assistant would have enough emotional intelligence to know when to gently correct its user, and when to hold back. At Google I/O 2025, Google showed a Gemini assistant that frequently corrected a user making obvious mistakes. But how long will it be before that kind of behaviour is seen as “bad UX”—the kind that pushes users away from the app, bringing retention and engagement metrics down? How long until the dark patterns we experienced on social media apps, like echo chambers, or new ones, come to haunt us, now supercharged by a personal AI assistant? What happens when your AI assistant is not only more responsive than your friends, but also more agreeable?
OpenAI is clearly steering toward a future where AI doesn’t just assist us—it understands us, shapes our routines, and perhaps even influences how we think and feel. And it’s not alone—Google, Meta, and others are close behind, each vying to become the default interface to the internet in the age of AI. The AI super assistant may well become one of the most transformative technologies ever built, redefining not only our relationship with computers but also our relationships with one another. If done right, it could make our digital lives more seamless, more supportive, and even more human. But as this technology becomes more deeply embedded in our lives, we must ask not just what it can do, but what it’s doing to us.
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
Neuralink raises $650 million Series E
Neuralink has raised $650 million in a Series E funding round, backed by major investors such as Sequoia Capital, ARK Invest, G42, and Founders Fund, to advance its brain–computer interface technology. The funding will support broader patient access and the development of next-generation devices aimed at restoring independence and expanding human potential. Last week, Semafor reported that this deal will value Neuralink at around $9 billion pre-money.
Coming to a Brain Near You: A Tiny Computer
The number of people with BCI (brain-computer interface) implants is expected to more than double in the coming year, driven by new FDA trial approvals and partnerships such as Apple’s collaboration with Synchron, reports The Wall Street Journal. The article highlights four leading BCI companies—Neuralink, Synchron, Precision Neuroscience, and Paradromics—each taking a distinct approach that differs in invasiveness, performance, and scalability. While mass adoption is a long-term goal, experts warn against overhyping consumer applications due to the risks of brain surgery and the experimental nature of the technology.
Neuralink competitor Paradromics completes first human implant
Neurotech startup Paradromics has successfully implanted its brain-computer interface in a human for the first time. The implant, called the Connexus Brain-Computer Interface, was inserted and removed from the patient’s brain in about 20 minutes during epilepsy surgery, demonstrating that it can be safely implanted and record neural activity. Though not yet FDA-approved, Paradromics plans to begin clinical trials later this year to study its long-term use, with the aim of helping patients with severe motor impairments communicate through a computer.
‘E-tattoo’ could track mental workload for people in high-stake jobs, study says
Researchers have developed a lightweight, flexible “e-tattoo” that can monitor brainwaves and eye movements to assess mental workload in real time. The device, made from conductive materials and worn on the forehead, performed as well as traditional EEG and EOG machines in tests and, when combined with a machine-learning algorithm, could alert users via an app when cognitive strain becomes too high, potentially helping professionals in high-stakes jobs like pilots and healthcare workers avoid errors and manage stress.
World's first therapy to reverse spinal cord injury enters human trial
The world's first regenerative cell therapy for spinal cord injury has been approved for a Phase I clinical trial by both the US FDA and China’s NMPA. Developed by Chinese biotech firm XellSmart, the therapy uses stem cells to potentially repair and regenerate damaged neural tissue, offering a scalable, off-the-shelf solution. This marks a major step toward treating a condition that affects over 15 million people globally and currently has no cure. Human trials are set to begin in partnership with a leading Chinese hospital.
🧠 Artificial Intelligence
Anthropic hits $3 billion in annualized revenue on business demand for AI
Anthropic has reached an estimated $3 billion in annualised revenue, tripling its December 2024 figure, according to Reuters, signalling rapid enterprise adoption of its AI services, particularly in code generation. Backed by Alphabet and Amazon, Anthropic is now among the fastest-growing SaaS companies ever, driven mainly by business clients rather than consumer users.
Introducing Bing Video Creator
Microsoft has launched Bing Video Creator, a free tool that lets users generate short AI-powered videos from text prompts, now available worldwide (except China and Russia) on the Bing Mobile App and coming soon to desktop. Powered by OpenAI’s Sora, it allows users to create 5-second videos in a 9:16 format. Microsoft says it has embedded safety measures to prevent misuse and mark AI-generated content.
AI pioneer Bengio launches $30M nonprofit to rethink safety
Yoshua Bengio, machine learning pioneer and one of the godfathers of deep learning, has launched a new non-profit lab, LawZero, aiming to develop AI systems that behave less like humans to avoid the risks of self-preserving, human-like machines. The new venture is backed by $30 million in funding, with initial backers including AI safety body the Future of Life Institute, Jaan Tallinn, a founding engineer of Skype, and Schmidt Sciences, a research body founded by former Google chief executive Eric Schmidt. Bengio argues that current training methods may lead to dangerous autonomy, and instead advocates for AI with intellectual detachment and strong oversight to prevent power concentration and misuse.
The latest Gemini 2.5 Pro is Google’s “most intelligent model yet”
Google has released an upgraded preview of its Gemini 2.5 Pro. According to benchmark results presented by Google, the new model shows improvements in coding, reasoning, and creativity. The model also features improved response structure and formatting based on user feedback. It is available immediately for developers through Google AI Studio, Vertex AI, and the Gemini app, with full rollout expected soon.
Meet Gemini in Chrome
Google continues its plan to Geminify everything by adding Gemini into Chrome. The new AI assistant will be able to summarise and answer questions about page content, extract information such as specifications or pros and cons, and assist with brainstorming and organising thoughts. Gemini in Chrome is currently available in early access for Google AI Pro and Ultra subscribers.
Claude Gov Models for U.S. National Security Customers
Anthropic has launched Claude Gov, a specialised set of AI models designed exclusively for US national security agencies, tailored for classified environments and tasks such as intelligence analysis, cybersecurity, and operational planning. They feature enhanced capabilities in handling sensitive materials and interpreting defence-related data. With this move, Anthropic follows OpenAI, which previously released ChatGPT Gov, a specialised version of its AI chatbot designed for US government agencies.
Musk Taps Investors for Billions Days After Washington Exit
According to Bloomberg, after stepping down from his advisory position to Donald Trump, Elon Musk is raising $5 billion in debt for his AI startup, xAI Corp. The funds, arranged by Morgan Stanley at high interest rates, will support projects such as a major data centre in Memphis as part of Musk’s aggressive expansion in artificial intelligence. According to the report, the offering has already attracted over $3.5 billion in demand. This move follows a broader fundraising spree across Musk’s companies, including Neuralink and a secondary xAI stock sale.
Cursor’s Anysphere nabs $9.9B valuation, soars past $500M ARR
Anysphere, creator of the AI coding assistant Cursor, has raised $900 million at a $9.9 billion valuation. The three-year-old startup has seen explosive growth, surpassing $500 million in annual recurring revenue—up 60% since April—and doubling its ARR roughly every two months. Anysphere reportedly declined acquisition offers from OpenAI and others earlier this year.
Nvidia’s Blackwell Conquers Largest LLM Training Benchmark
Nvidia has once again dominated the MLPerf training benchmarks, with its Blackwell GPUs securing top positions across all six machine learning tasks. However, systems built with AMD’s MI325X GPU achieved results on par with Nvidia H200 GPUs, suggesting that AMD is now just one generation behind. Notably, Nvidia’s NVL72 system achieved near-linear performance scaling across thousands of GPUs, underscoring the growing importance of efficient networking. In terms of power consumption, only Lenovo submitted data, highlighting the need for greater transparency in future benchmark rounds.
DeepSeek may have used Google’s Gemini to train its latest model
According to one researcher, DeepSeek’s latest AI model, R1-0528, may have been partially trained on data from Google’s Gemini, based on linguistic similarities and trace patterns. While not definitive proof, the claims have reignited scrutiny over DeepSeek’s practices, including past allegations of training on outputs from OpenAI’s ChatGPT. Experts note that AI models increasingly exhibit overlapping language due to web “contamination” from AI-generated content. Meanwhile, firms like OpenAI and Google are tightening security to prevent unauthorised data distillation by rivals.
FutureHouse previews an AI tool for ‘data-driven’ biology discovery
FutureHouse, a non-profit backed by Eric Schmidt, has unveiled Finch, a new AI tool designed to assist biological research by analysing scientific literature and generating data-driven insights. Marketed as akin to a first-year graduate student, Finch automates parts of the scientific process, though it remains in closed beta and is still prone to making errors.
AI could account for nearly half of datacentre power usage ‘by end of year’
According to a new report, AI systems could consume nearly half of all data centre electricity by the end of 2025. The report suggests that AI could require up to 23 gigawatts of power, double the energy used by the Netherlands. This comes as the International Energy Agency predicts AI will need as much energy by 2030 as Japan uses today.
How Anthropic Is Snatching Top Talent from OpenAI and DeepMind
The AI sector is in the midst of an intense talent war, with companies like Anthropic, OpenAI, DeepMind, and major tech firms fiercely competing for elite researchers and engineers through multimillion-dollar compensation packages, equity incentives, and even aggressive poaching tactics. Anthropic has emerged as a top destination, particularly for former OpenAI and DeepMind staff, and boasts the highest retention rate among AI labs.
🤖 Robotics
Amazon ‘testing humanoid robots to deliver packages’
Amazon is developing AI software for humanoid robots intended to deliver packages either alongside human drivers or independently, according to a report published by The Information. A “humanoid park” has reportedly been established at an Amazon office in San Francisco to test the robots on an indoor obstacle course. The aim is for the robots, built by external companies, to operate from Amazon’s Rivian vans, potentially delivering packages while human drivers serve other addresses. Following trials in controlled environments, Amazon plans to deploy the robots on “field trips” to attempt real-world home deliveries.
Tesla’s Head of Optimus Humanoid Robot Program Exits Company
Milan Kovac, head of engineering for Tesla’s Optimus humanoid robot project, has announced that he is leaving the company effective immediately. Leadership of Optimus will shift to Ashok Elluswamy, head of Tesla's Autopilot division. In his post on X, Kovac cited a desire to prioritise family as the reason for leaving and expressed continued support for the team.
Amazon delivery drones crashed after mistaking rain for ground
A recent Bloomberg report has revealed that a software update caused Amazon’s MK30 drones to crash during December test flights in Oregon, forcing a temporary pause in the company's drone delivery operations. The update made the drones overly sensitive to rain, leading them to shut down mid-air mistakenly. Experts say the removal of mechanical "squat switches" from the MK30 contributed to the crashes, as it left the drones reliant solely on software to detect landings. Amazon has disputed Bloomberg’s conclusions but has since updated its systems and resumed testing in March.
Who Gives a S#!t About Cursing Robots?
Researchers from Oregon State University asked an important question: Should robots curse? Although the initial response might be that they shouldn’t, the team found that robots using profanity when making mistakes were often seen as more humorous and socially engaging. While reactions from the general public were more mixed, finding swearing robots slightly more discomforting, acknowledging errors mattered more than the language used. The researchers argue that, with thoughtful implementation, cursing robots could challenge outdated social norms and enhance human-robot interaction.
🧬 Biotechnology
Biotech IPO Slump Seen Dragging Into 2026 as Returns Disappoint
Biotech IPO activity in the US remains sluggish, with only five companies raising over $50 million going public so far this year, as the sector continues to underperform amid investor caution, regulatory uncertainty, and political headwinds. The Nasdaq Biotechnology Index has lagged broader markets, reflecting poor returns from many companies that went public during the pandemic. Analysts note that while long-term prospects remain strong due to scientific advances and big pharma’s need to replenish pipelines, a full rebound in biotech listings is not expected until 2026, with investors now favouring more mature, later-stage firms.
Nobel Prize winners convince court to revive CRISPR patent dispute
A US appeals court has revived the University of California and the University of Vienna's challenge for patent rights to CRISPR gene-editing technology in a long-running dispute with the Broad Institute. The Federal Circuit ruled that the US Patent Office had misapplied patent law in favour of the Broad Institute and ordered a reassessment, potentially overturning previous decisions and allowing the universities to press their case as the original creators of CRISPR.
Human Brain Cells on a Chip for Sale
Australian startup Cortical Labs has launched the CL1, the world’s first commercially available biological computer powered by lab-grown human brain cells. Combining live neurons with silicon hardware, the CL1 can process information in real time and adapt to stimuli, offering researchers a platform for studying brain function, drug discovery, and disease modelling. With each unit containing 800,000 neurons and priced at $35,000, the system has attracted interest from academia, biotech, and even the arts, marking a major step forward in neuromorphic and bioengineered computing.
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!"