Hello and welcome to Sync #509!

The main story of this week’s issue is Apple Intelligence—or, more specifically, the lack of it. We take a closer look at the recently announced delay and what it means for Apple’s ambitious foray into artificial intelligence.

Elsewhere in AI, Anthropic has raised another $3.5 billion in Series E funding at a $61.5 billion valuation and submitted its AI policy recommendations to the White House—while quietly removing Biden-era AI policy commitments from its website. Meanwhile, a judge has expedited the Musk vs OpenAI trial for later this year. We also have AI agents communicating in an efficient robotic language, DeepSeek publishing statistics on its online service, and a new report warning of the escalating risks posed by rapid advancements in artificial intelligence.

Over in robotics, Clone Robotics presents its android powered by artificial muscles, while the Unitree G1 humanoid robot shows off its dance moves and why the future of robotics isn’t necessarily humanoid.

Additionally, this week’s issue of Sync features genetically engineered woolly mice, highlights from the ARDD Emerging Science & Technologies Workshop, how to create superbabies, and more.

Enjoy!

Where’s Apple Intelligence?

Every new product Apple released in the past year was advertised as “built for Apple Intelligence”—from iPhones and iPads to Macs. Apple Stores have been remodelled to highlight Apple Intelligence. Billboards for the latest iPhones say, “Hello, Apple Intelligence.” However, the company finds itself struggling to deliver on its promises made at WWDC 2024, and Apple’s ambitious foray into artificial intelligence is facing significant roadblocks.

While Apple Intelligence was heavily marketed as a core selling point of the iPhone 16, many of its most promising features are still missing or incomplete. Rather than rolling out Apple Intelligence as a single update, Apple has drip-fed features into its ecosystem. This staggered release strategy has left consumers confused and disappointed. The AI features currently available—such as voicemail transcriptions, Genmoji, image generation, and Writing Tools—are seen as minor conveniences rather than revolutionary advancements. Visual Intelligence, which uses the camera to provide information on items in the real world, feels like a more basic version of Google Lens. Worse still, some features, like notification summaries, have proven unreliable, generating inaccurate or misleading information and forcing Apple to disable them.

The biggest disappointment is Siri’s highly anticipated overhaul. Or, to be more precise, the lack of it. Apple recently confirmed that the new version of its voice assistant, which was supposed to be deeply integrated with Apple Intelligence and Apple’s ecosystem, will not launch in iOS 19 or macOS 16 this year. Given Apple’s usual software release cycle, this means the revamped Siri may not debut until autumn 2026 or even 2027.

Meanwhile, competitors are steaming ahead. Amazon recently revealed the new Alexa+, bringing a much-needed refresh and a modern conversational AI to its voice assistant. The latest Samsung smartphones, such as the S25 Ultra, feature integration with Google’s Gemini assistant. Then there are the companies that emerged from the generative AI boom, like OpenAI and Anthropic, which have set new standards for conversational AI, further highlighting the gap between Apple and its competitors.

With OpenAI, Google, and Amazon launching more advanced AI assistants, Apple’s slower AI adoption risks alienating users who expect more from their premium devices. The failure to release a fully revamped Siri could hurt sales of Apple’s next-generation iPhones and Macs, especially as rivals offer AI-driven products that feel more futuristic and intuitive. In a recent earnings call, Tim Cook admitted that in the markets where Apple rolled out Apple Intelligence, the year-over-year performance of the iPhone 16 family was stronger than in those where Apple Intelligence was not available. However, many of those early iPhone 16 buyers may feel misled, as the AI features Apple promised were not ready at launch. If Siri and other Apple Intelligence features continue to underdeliver, Apple risks eroding consumer trust in its brand.

Apple’s struggles to deliver what it promised at WWDC 2024 reflect its internal challenges, including leadership issues, talent retention, and technical bottlenecks. As Mark Gurman reports, the company has been working on a next-generation “LLM Siri” architecture, designed to modernise Siri using large language models. However, development has been slower than expected, with reports suggesting that Apple’s fragmented AI infrastructure has made integration of legacy Siri with LLM Siri difficult. Engineers have been forced to balance maintaining the legacy Siri system while simultaneously developing a new AI-powered version, leading to delays and inefficiencies. The new architecture is expected to unify these systems, but it may not be fully realised until iOS 20 or later, potentially pushing back Apple’s AI roadmap beyond 2026. Adding to these struggles, Apple has reportedly lost key AI talent to competitors like OpenAI and Google DeepMind, further slowing down progress. With industry-wide demand for AI experts at an all-time high, Apple’s inability to retain top engineers may be further exacerbating delays.

Another reason for Apple’s slow AI rollout is its commitment to privacy. Unlike Google and OpenAI, which rely on vast amounts of user data to improve their AI models, Apple maintains a strict stance on on-device processing and data security. Apple even makes its focus on privacy a key selling point of Apple Intelligence. While this focus on privacy is admirable, it also limits Apple’s ability to develop AI at the same pace as its competitors.

Apple still has time to deliver something great, but the window is closing fast. Apple Intelligence could be an industry-leading AI ecosystem given Apple's control over hardware, software, and services—all with a big focus on privacy. If the company finally launches the overhauled Siri in 2026 or 2027, it will need to ensure it is not just playing catch-up but offering something truly innovative. Otherwise, Apple risks losing ground in one of the most transformative technological shifts in recent history. For now, Apple Intelligence remains more of a marketing slogan than a promised revolution in how we interact with computers.

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

Share

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

Become a paid subscriber

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

Innovations in aging biology: highlights from the ARDD emerging science & technologies workshop
This paper highlights the talks presented at the Emerging Science & Tech Workshop during the 2024 AARD conference, each offering different perspectives on accelerating discoveries in ageing science, from AI and advanced imaging to neurotechnology, synthetic biology, and cryopreservation. The workshop concluded that advancing ageing research requires a systems-based, high-throughput approach rather than linear models, emphasising the need to integrate AI, neurotechnology, and synthetic biology. Experts highlighted the underutilisation of existing technologies, such as genome sequencing and advanced imaging, and called for greater collaboration to accelerate breakthroughs in longevity science.

How to Make Superbabies
This article presents a controversial yet technically grounded argument for using gene editing to create “superbabies” with enhanced intelligence, longevity, and health while criticising the scientific community's reluctance to explore human enhancement due to ethical concerns. While embryo selection offers limited improvements, gene editing—particularly through CRISPR and stem cell-derived embryos (Super-SOX)—could unlock significant genetic gains, including an IQ boost of 50+ points and a life expectancy increase of 15–50 years. Additionally, the author argues that superbabies could serve as a backup plan against uncontrolled AI development by creating hyper-intelligent humans capable of addressing AI alignment and existential risks. The article calls for investment, policy reform, and breaking the societal taboo around human genetic enhancements.

A Self-Balancing Exoskeleton Strides Toward Market

Thanks to the XoMotion Exoskeleton, a powered exoskeleton, Chloë Angus—who lost sensation in her legs a decade ago due to a benign spinal cord tumour—was able to dance at CES 2025. Created by the Canadian company Human in Motion Robotics, this self-balancing exoskeleton eliminates the need for crutches or walkers, aiming to restore mobility for people with spinal cord injuries. Clinical trials are set to begin in April 2025, initially in rehabilitation facilities, with plans for a home-use model in the future.

🔮 Future visions

▶️ Methuselah Civilizations: A Society of the Ageless (36:10)

In this video, Isaac Arthur explores the idea of Methuselah civilisations—hypothetical civilisations that achieved extreme longevity, either biologically or through technological means. Arthur imagines how such civilisations might function and examines their cultural, psychological, and economic implications. As he speculates on life within a Methuselah civilisation, he asks what would drive progress, ambition, and purpose in a world where time has no limits.

🧠 Artificial Intelligence

Anthropic raises Series E at $61.5B post-money valuation
Anthropic has raised $3.5 billion in Series E funding at a $61.5 billion valuation, with backing from investors such as Lightspeed Venture Partners, Bessemer, Cisco, and Fidelity. In total, the company has raised $18.2 billion to date. This funding will support the development of next-generation AI systems, the expansion of compute capacity, and research into AI interpretability and alignment, while also accelerating global growth. Anthropic recently launched Claude 3.7 Sonnet and Claude Code, focusing on advanced coding capabilities and AI collaboration. Additionally, Claude now powers Alexa+, bringing enhanced AI features to millions of Amazon Prime users.

T-Mobile’s parent company is making an AI Phone with Perplexity
Deutsche Telekom (T-Mobile's parent company) has announced an “AI Phone,” which will be developed in partnership with Perplexity. Similar to Google’s Gemini on Pixel or Apple’s Siri, Perplexity AI will be integrated at a system-wide level, including from the Lock Screen. Additionally, the AI Phone will feature Google Cloud AI, Picsart, and ElevenLabs apps, as well as potential agentic capabilities. The phone is expected to cost below $1,000 and is set to be officially launched in the second half of 2025, with a release date expected sometime in 2026.

Superintelligence Strategy
Superintelligence Strategy is a report written by Dan Hendrycks (director of the Center for AI Safety), Eric Schmidt (former CEO and chairman of Google), and Alexandr Wang (founder and CEO of Scale AI), warning of the escalating risks posed by rapid artificial intelligence advancements. The authors argue that superintelligent AI could destabilise global security, much like nuclear weapons did in the past. They propose a three-pronged strategy of deterrence, nonproliferation, and competitiveness to manage AI’s risks while harnessing its benefits. Central to their framework is "Mutual Assured AI Malfunction" (MAIM), a deterrence model akin to nuclear MAD, where nations may sabotage rival AI projects to prevent unilateral dominance. They also stress the need for strict controls on AI technology to prevent misuse by rogue actors and advocate for domestic AI investments to ensure economic and military resilience. The report urges policymakers to act pragmatically, warning that failure to regulate AI’s development could lead to catastrophic outcomes or an unchecked superintelligence arms race.

▶️ Two AI agents on a phone call realize they’re both AI and switch to a superior audio signal ggwave (1:09)

Human language may not be the most efficient tool for AI agents to communicate. At the ElevenLabs 2025 Hackathon in London, a team of developers created gibberlink, a robotic-sounding language designed for more effective AI communication. The creators of gibberlink claim that this method is significantly cheaper, as it does not require a GPU for voice synthesis or speech recognition and can run entirely on a CPU. You can try out gibberlink yourself here. Additionally, the project is open-source and available on GitHub.

Musk’s Fight With OpenAI Set for Expedited Trial This Year
The judge in the Elon Musk vs. OpenAI case, which centres on whether OpenAI’s transition from a nonprofit to a for-profit company is unlawful, has expedited the lawsuit and set a trial for the autumn of this year. However, the judge denied Musk’s request to immediately halt OpenAI’s restructuring while acknowledging the public interest and potential harm if the transition is found to be illegal. OpenAI welcomed the rejection of the injunction, arguing that Musk’s lawsuit is driven by competition rather than principle. Meanwhile, OpenAI is working to secure regulatory approval for its for-profit shift from authorities in Delaware and California.

Anthropic submits AI policy recommendations to the White House
Anthropic has submitted a set of AI policy recommendations to the White House, advocating measures to strengthen national security, energy infrastructure, and AI adoption across government agencies. Among its proposals, Anthropic calls for maintaining the AI Safety Institute, directing NIST to develop national security evaluations, and creating a government team to analyse AI security vulnerabilities. The company also urges tighter AI chip export controls, particularly restricting Nvidia H20 chip sales to China, and setting a national target of 50 additional gigawatts of power for AI development by 2027. Interestingly, the move comes just a day after the company quietly removed Biden-era AI policy commitments from its website.

Amazon is reportedly developing its own AI ‘reasoning’ model
Amazon is reportedly joining the reasoning AI race with its own reasoning model. The new model could be unveiled as soon as June under the Amazon Nova brand and may feature a hybrid approach, similar to the recently released Claude 3.7 Sonnet.

Did xAI lie about Grok 3’s benchmarks?
There was some drama on X last week over Grok 3’s benchmarks. An OpenAI employee accused xAI of publishing misleading benchmark results for its latest AI model. xAI claimed that Grok 3 outperformed OpenAI’s o3-mini-high on the AIME 2025 math benchmark, but OpenAI employees pointed out that xAI had omitted certain scores that would place Grok 3 behind o3-mini-high. However, as this article highlights, both companies engage in benchmark hacking to present their models in the best possible light, emphasising that the real missing metric is the computational and financial cost required to achieve top scores.

Statistics of DeepSeek's Online Service
On the sixth day of its Open-Source Week, DeepSeek published a document detailing the design of its inference system, offering insights into how its engineers addressed the challenge of serving V3 and R1 models with high throughput and low latency. Interestingly, in the final section, DeepSeek also shared usage data, revealing that if all tokens were billed at DeepSeek-R1’s pricing, the total daily revenue would be $562,027, with a cost profit margin of 545%. However, the actual revenue is significantly lower.

How DeepSeek became a fortune teller for China’s youth
DeepSeek, China’s advanced AI reasoning model, has sparked a massive trend in AI-driven fortune-telling, with over two million posts in February alone mentioning “DeepSeek fortune-telling” on WeChat, the country’s largest social platform. Unlike traditional fortune tellers, DeepSeek is free, available 24/7, and widely considered more accurate than ChatGPT in BaZi readings, making it an appealing spiritual tool for many young Chinese users. However, while it effectively processes life-chart information, some argue that it lacks the human intuition needed for deeply personalised readings. Additionally, researchers warn of potential psychological risks, as overly deterministic AI predictions could lead to false hope or unnecessary fear.

The Deep Research problem
In this post, Benedict Evans shares his experience with Deep Research, OpenAI’s AI tool designed to automate research and data analysis. At first glance, it seemed perfectly suited for his work, promising to save hours of effort. However, in practice, it fails in critical ways, producing errors that undermine its reliability. Evans highlights how the tool misinterprets sources, retrieves incorrect data, and lacks precision, making it untrustworthy for serious research. While AI can be a powerful assistant, Deep Research’s shortcomings suggest that it is not yet a replacement for human expertise.

▶️ What Is the Most Popular Open-Source AI Stack? (4:19)

In this video, ByteByteGo showcases a range of open-source projects and tools for building an AI app, covering everything from frontend tools to embeddings, RAG libraries, and open large language models that together form the emerging and constantly evolving open-source AI stack.

SWE-Lancer: Can Frontier LLMs Earn $1 Million from Real-World Freelance Software Engineering?
Researchers from OpenAI present SWE-Lancer—a new benchmark assessing AI models on their abilities to solve real-life coding tasks. The benchmark contains 1,400 freelance software engineering tasks from Upwork, valued at $1 million total in real-world payouts. As reported in the paper, no model has yet been able to claim the full $1 million payout, with the best-performing model, Claude 3.5 Sonnet, achieving 40% when tested against the full SWE-Lancer dataset, which includes both coding and managerial tasks.

If you're enjoying the insights and perspectives shared in the Humanity Redefined newsletter, why not spread the word?

Refer a friend

🤖 Robotics

▶️ Protoclone: Bipedal Musculoskeletal Android V1 (0:40)

This android from Clone Robotics is an interesting humanoid robot, to say the least. While other humanoid robots, such as Figure 02, Boston Dynamics’ Atlas, Unitree H1, Tesla Optimus, and others, use electric motors to move their limbs, Protoclone uses artificial muscles. According to Clone, this unique approach allows for the creation of an anatomically accurate synthetic human with over 200 degrees of freedom, resulting in an amazing feat of engineering—or nightmare fuel, depending on your viewpoint.

▶️ Keep the Music Going, Keep the Dance Flowing! (0:21)

In this video, Unitree showcases the dancing skills of its humanoid robot while it is being hit with a stick and distracted in various other ways. The robot remains unfazed and continues to demonstrate its dexterity, body control, and coordination.

Apptronik’s humanoid robots take the first steps toward building themselves
Apptronik has announced a pilot partnership with Jabil, a major supply chain and manufacturing company. This is Apptronik’s second pilot partnership—the company previously partnered with Mercedes-Benz in March 2024, though that partnership is still in the pilot stage. Additionally, Jabil has become Apptronik’s manufacturing partner, opening the possibility that, if the pilot is successful, Jabil will produce Apollo, Apptronik’s humanoid robots, in its own factories, potentially leading to robots manufacturing themselves. Apptronik targets 2026 for commercial production and is also collaborating with Google DeepMind to develop AI for its humanoids.

Why the future of robotics isn’t necessarily humanoid
The robotics industry is fascinated by humanoid robots, but they aren’t necessarily the best solution for real-world automation challenges. In some industries, such as warehouse logistics, efficiency should take precedence over making robots resemble humans, this article argues. Just as nature does not force all species into a human shape, robots should also be optimised for function, and the human form is not always the best form. While research in humanoid robotics will drive technological advancements, the article states that current industry needs are better served by specialised robots.

Humanoid robots can swiftly get up after they fall with new learning framework
Developed by researchers at the University of Illinois Urbana-Champaign, HUMANUP is a new framework that uses reinforcement learning to enable robots to autonomously recover from falls—albeit in a very uncanny way. When tested on the Unitree G1 humanoid robot in both simulations and real-world environments, HUMANUP succeeded in 78.3% of cases compared to 41.7% for the built-in controller. Additionally, robots equipped with HUMANUP successfully recovered from face-up and face-down falls on various surfaces, including slippery snow, grass, and slopes.

Morphing robot turns challenging terrain to its advantage

The GOAT (Good Over All Terrains—great name, by the way) is a robot inspired by animals that adapt to their environment, such as mountain goats and armadillos. GOAT can switch between a flat ‘rover’ shape and a sphere, enabling it to drive, roll, and even swim efficiently. Designed for flexibility and adaptation, the robot features a frame made of elastic fibreglass rods and rimless motorised wheels. Unlike traditional robots that compute the shortest path, GOAT considers the best travel modality—such as rolling downhill or swimming through obstacles—to save time and energy. The researchers behind the robot hope it will be useful for environmental monitoring, disaster response, and possibly even extraterrestrial exploration.

🧬 Biotechnology

Scientists aiming to bring back woolly mammoth create woolly mice
Colossal Biosciences is a biotech company with an ambitious mission—to bring back woolly mammoths by genetically modifying Asian elephants to acquire mammoth traits, with the first calf hoped to be born by 2028. This week, the company took a step towards that goal by announcing the successful creation of genetically modified mice with woolly hair traits. Researchers altered nine genes linked to hair characteristics and cold adaptation, resulting in some mice developing woolly coats, long hair, or golden-brown fur. A gene related to fat metabolism was also modified but did not significantly affect body mass. Colossal Biosciences plans to conduct behavioural tests to assess the modified mice's cold tolerance. However, bringing back woolly mammoths will be a far greater challenge and further research is needed to determine how these findings apply to elephants.

With generative AI, MIT chemists quickly calculate 3D genomic structures
MIT chemists have developed a new generative AI model to predict 3D genome structures significantly faster than existing experimental techniques. The model, called ChromoGen, integrates deep learning and generative AI to analyse DNA sequences and predict chromatin conformations with high accuracy. Unlike traditional experimental methods such as Hi-C, which are time-consuming and labour-intensive, ChromoGen can generate thousands of structures in minutes. When tested against experimental data, the model produced highly similar 3D structures ChromoGen is open-source and available on GitHub.

💡Tangents

Firefly Aerospace Becomes First Commercial Company to Successfully Land on the Moon

Blue Ghost has landed on the Moon, making Firefly Aerospace the first commercial company in history to achieve a successful soft landing on the lunar surface. Blue Ghost is now conducting surface operations and supporting several NASA science and technology demonstrations over the next 14 days, including lunar subsurface drilling, sample collection, X-ray imaging, and dust mitigation experiments. It will also capture photos of a total eclipse when the Earth blocks the Sun above the Moon’s horizon. Additionally, Blue Ghost has become the first spacecraft to use GPS signals on the Moon, marking a significant step for future lunar missions.

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

Share

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!"

Message Conrad Gray