The Dire Wolves are back… or are they? - Sync #514

Plus: 2025 AI Index Report; Kawasaki rideable robot-wolf concept; Shopify memo on AI; Google's TPU for the age of inference; DeepMind's AI master Minecraft; and more!

Hello and welcome to Sync #514!

This week, Colossal Biosciences proclaimed that dire wolves are back and unveiled three white-furred pups the company claims have been brought back from extinction. We’ll take a closer look at these claims, the reactions they’ve sparked, and whether this marks a genuine scientific breakthrough or a well-timed publicity stunt.

In other news, the 2025 AI Index Report is out, offering a snapshot of the progress AI has made over the past year across research and industry. Elsewhere in AI, a leaked Shopify memo reveals its AI-first strategy, DeepMind’s AI has learned to play Minecraft, Google has announced new TPUs for the age of inference, and former OpenAI employees are raising billions for their new AI startups.

In robotics, Kawasaki has revealed a futuristic concept for a rideable, all-terrain four-legged robot, due to be completed in around 25 years. Meanwhile, Waymo is preparing its autonomous cars for their first international tests in Japan, Samsung has released a home robot, and a humanoid robot made a catwalk debut at Shanghai Fashion Week.

Also featured in this week’s issue of Sync: the world’s smallest pacemaker (smaller than a grain of rice), why brain-computer interfaces face a critical test this year, TSMC’s most advanced chip manufacturing technology, and more!

Enjoy!

The Dire Wolves are back… or are they?

Image generated with Midjourney

After over 10,000 years, the eerie howl of a long-extinct predator is echoing once again—at least, that’s what Colossal Biosciences would have you believe. In a secretive preserve somewhere in the United States, three snow-white wolf pups named Romulus, Remus, and Khaleesi are being hailed as the first “de-extincted” dire wolves. These animals represent what their creators claim is the world’s first successful reversal of extinction.

The announcement ignited a media frenzy. But while Time magazine featured the pups on its cover and Game of Thrones author G.R.R. Martin took a photo with them, the discussion erupted, asking how these dire wolves were made, whether we should even be doing it, why Colossal Biosciences did it, and whether those dire wolves are actually dire wolves.

But are these wolves truly back from extinction—or are they something new entirely? What does it mean to “resurrect” a species in the age of gene editing? Are we witnessing a breakthrough in conservation science, or a high-tech publicity stunt?

What Is Colossal Biosciences?

Before we explore how these dire wolves have been brought back from extinction, let’s take a closer look at the company behind them—Colossal Biosciences. Colossal Biosciences was founded in 2021 by tech entrepreneur Ben Lamm and renowned Harvard geneticist George Church, with a vision as bold as it is controversial: to reverse extinction. Branding itself as “the world’s first de-extinction company,” Colossal isn’t just in the business of conservation—it’s aiming to rewrite the rules of life on Earth.

The company’s mission goes far beyond the dire wolf. Its de-extinction wish list includes the woolly mammoth, the dodo, and the thylacine, or Tasmanian tiger. Using genome sequencing, CRISPR-style gene editing, and cloning, Colossal aims to resurrect lost species not by simply copying old DNA, but by engineering modern relatives to express the traits of their extinct counterparts.

This synthetic biology approach allows Colossal to frame de-extinction not just as a scientific marvel, but as a tool for environmental repair. The company argues that bringing back extinct species could help restore damaged ecosystems—by reintroducing lost predators or grazing patterns—and even help endangered species survive climate change by editing their genomes for greater resilience.

Colossal has already raised almost $450 million from investors, including celebrity backers like Chris Hemsworth, Tom Brady, Peter Jackson, and George R.R. Martin. With a current valuation of $10.2 billion, the company is as much a high-profile biotech juggernaut as it is a scientific venture.

About a month ago, Colossal grabbed some news headlines with its genetically modified woolly mice. But its most ambitious—and most contested—project yet is the one that’s captured attention and interest around the world: the return of the dire wolf.

What Colossal Biosciences Actually Did to Create Those Dire Wolves

The story of Romulus, Remus, and Khaleesi begins in the lab. Far from reviving a perfectly preserved dire wolf from the Ice Age, Colossal’s team used a combination of ancient DNA analysis, genome editing, and cloning to produce wolves that resemble the extinct species—but are not literal copies.

To start, Colossal scientists extracted DNA from two dire wolf fossils: a 13,000-year-old tooth found in Ohio and a 72,000-year-old ear bone from Idaho. These ancient samples provided enough genomic information to build a partial genetic map of the long-extinct Canis dirus. Using that data, Colossal identified 14 genes—across which they made 20 specific edits—that appeared to influence the dire wolf’s most recognisable traits: a more massive frame, a broader skull, stronger jaws, thick fur, and a light-coloured coat.

But instead of inserting ancient DNA directly, the team employed a technique known as multiplex base editing to rewrite the genes of a living relative: the grey wolf. Using cells taken from a modern grey wolf, scientists altered the DNA to reflect the dire wolf’s genetic profile. These edited cells were then used in somatic cell nuclear transfer—a cloning process where the nucleus of an edited cell is implanted into an egg cell (from which the original DNA has been removed).

The resulting embryos were implanted into large domestic dogs, chosen for their ability to safely carry the sizeable pups. Out of dozens of embryo transfers, three births were successful: Romulus and Remus in October 2024, and Khaleesi in January 2025. A fourth pup died from an infection shortly after birth—a reminder that cloning is still an imperfect and often fragile process.

Since their birth, the wolves have been raised in a 2,000-acre private preserve under round-the-clock observation and tight security. They were initially fed by hand, weaned on a diet of puréed meat, and eventually transitioned to tearing apart whole cuts of deer and beef. Early on, they exhibited behaviour consistent with wild instincts: wariness around humans, early howling, and even “stalking” leaves blown by the wind.

Are These Actual Dire Wolves?

While the science behind their creation is undeniably complex and groundbreaking, it leads to an unavoidable question: Are these really dire wolves?

Colossal’s wolves may look the part, but whether they are dire wolves remains hotly contested. On a biological level, dire wolves (Canis dirus) and grey wolves (Canis lupus) share some superficial similarities—both are large, social canids adapted to hunting in packs. But genetically, they are distant cousins, having diverged around 5.7 to 6 million years ago. In fact, modern jackals are more closely related to grey wolves than dire wolves are.

This genetic distance is what has many scientists pushing back against Colossal’s claim. The pups are the result of 20 gene edits—targeting traits like skull width, jaw strength, body size, fur colour, and vocalisations—out of approximately 19,000 total genes in the grey wolf genome. While these edits produce animals with some dire wolf-like features, they fall far short of recreating the full genetic identity of the extinct species.

Colossal’s wolves aren’t pure grey wolves anymore. For some, they are more like GMO grey wolves. But they are also not true dire wolves, reconstructed in full from ancient DNA. They are, in essence, a genetically sculpted facsimile—an approximation of what we think a dire wolf looked and behaved like, raising the question of whether resemblance is enough to claim revival.

Beth Shapiro, Colossal’s Chief Science Officer, offers a functionalist view, saying that “if it looks like that species, if it is acting like that species, if it’s filling the role of that species, then you’ve done it.” Others, like evolutionary biologists and taxonomists, argue that species identity is more than skin deep—or even gene deep. It’s about evolutionary lineage, reproductive continuity, and ecological history.

De-Extinction in Service of the Living

Colossal frames these de-extinction projects as technological stepping stones towards protecting currently endangered species. In parallel with the dire wolf project, Colossal has successfully cloned four American red wolves, a critically endangered species with fewer than 20 individuals remaining in the wild. These clones, derived from canids carrying rare red wolf “ghost alleles,” could help restore genetic diversity and rescue the species from a severe bottleneck.

Similarly, Colossal is applying its techniques to species like the northern white rhino—of which only two are left—by reconstructing lost diversity from preserved tissue samples. And in Australia, the company is engineering “super quolls” to survive the invasive cane toad, whose toxins have devastated native predators. By identifying and copying resistance genes from South American species, Colossal hopes to make the quoll population resilient enough to recover.

Supporters of de-extinction and Colossal’s approach argue that we need every tool available to preserve biodiversity, including synthetic biology. Detractors warn that this line of thinking could distract from simpler, proven conservation measures: protecting habitats, enforcing anti-poaching laws, and curbing climate change.

Scientific Breakthrough or PR Stunt?

Colossal Biosciences’ dire wolf project occupies a curious space—somewhere between legitimate scientific innovation and media spectacle. On one hand, the company’s successful use of multiplex gene editing, cloning, and ancient DNA analysis represents a real milestone in synthetic biology. Very few labs have ever produced healthy animals with this many genetic edits, and fewer still have done it using extinct genome reconstructions as their blueprint.

Still, it’s hard not to notice the media campaign that followed: a front cover story in Time magazine, an interview with Joe Rogan, involving G.R.R. Martin, an entire series of videos dropped on Colossal’s official YouTube channel. The campaign definitely brought a lot of eyes to the company. However, what was missing in this campaign was a scientific paper for independent verification. Apparently, a scientific paper is forthcoming on the bioRxiv preprint server—but as of now, independent verification remains pending.

The truth, perhaps, is that it’s both.

Colossal has undeniably advanced the tools of conservation genetics. Its work could help save species like the red wolf, the northern white rhino, and the quoll. But by framing genetically modified wolves as “de-extincted,” the company has also blurred the line between restoration and reinvention. Romulus, Remus, and Khaleesi may not be true dire wolves—but they are symbols of a future in which humans no longer just shape the environment—we reshape life itself.

Whether that future is one of healing or hubris remains to be seen.

---

De-extinction is a fascinating, complex topic—equal parts science, philosophy, and speculation. There’s far more to unpack than can fit in a single article. From the mechanics of ancient DNA editing to the ecological ripple effects of reintroducing extinct species, this subject deserves a deeper dive—or perhaps an entire series.

That’s exactly what’s been in the works for a couple of weeks now, so stay tuned for when those articles are finally out!

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

Brain-computer interfaces face a critical test
Implanted brain-computer interfaces (BCIs), once a niche experiment, are now gaining traction as clinical trials expand and more companies enter the field. The next 5–10 years are seen as critical: BCIs could either become a widely adopted assistive technology or remain confined to experimental research. Despite ongoing technical and regulatory challenges, researchers remain optimistic that a breakthrough moment is on the horizon. If you want to learn more about BCIs, check out my Brain-Computer Interfaces series.

World’s Tiniest Pacemaker Is Smaller Than a Grain of Rice

Researchers have created the world’s smallest pacemaker—smaller than a grain of rice—that can be implanted through a stent and controlled wirelessly using infrared light. Additionally, the device safely dissolves in the body after use. Its tiny size allows for minimally invasive use, especially benefiting infants and post-surgery patients. Beyond cardiac care, the technology could be adapted for nerve regeneration, pain management, and wound healing.

🧠 Artificial Intelligence

The 2025 AI Index Report
Stanford HAI has released its 2025 AI Index, a snapshot of the progress AI has made in the last year across research and industry. This year, the report highlights major gains in AI capabilities, record-breaking private investment ($109.1B in private AI investment in the US), more efficient, affordable and accessible models, and growing integration into sectors like healthcare and transportation. While optimism about AI is rising globally (especially in Asia), the Index also underscores persistent challenges in responsible AI practices, educational access, and complex reasoning tasks. Governments are stepping up with new regulations and funding, as AI becomes increasingly central to science, business and daily life. Notably, China is closing the performance gap with the US, leading in AI publications and patents while expanding access to autonomous services and making major public investments in AI infrastructure. The full, 456-page report is available here. Additionally, the team behind the report has also prepared an easy-to-digest collection of 10 charts showing the state of AI and highlighting key takeaways.

Shopify CEO says before hiring anyone new, employees must prove AI can't do the job better
Shopify goes all in on AI. In a company-wide memo, Shopify CEO Tobi Lütke has mandated that employees must prove a task can't be accomplished with AI before requesting additional headcount or resources, signalling a major shift towards AI-driven operations. Additionally, Lütke declared AI usage a baseline expectation for all staff, with performance reviews now factoring in AI adoption.

Anthropic: Introducing the Max Plan
Anthropic follows OpenAI’s example and introduces Max Plan—a new premium pricing tier for Claude power users who need significantly higher usage limits. The Max Plan comes in two levels—$100/month gives 5x more usage than the Pro plan, and $200/month gives 20x usage than the Pro plan. Additionally, customers on Max Plan will have priority access to the newest features and models.

Midjourney V7 Alpha
Midjourney has launched an alpha version of its new V7 model for community testing, which introduces major upgrades in image quality, prompt comprehension, and coherence in details like hands and objects. Key features include default-enabled personalisation and a new "Draft Mode," which renders images at 10× speed and half the cost, allowing for quick, conversational-style iterations.

Alibaba’s Tsai Warns of ‘Bubble’ in AI Data Center Buildout
Tech giants across the globe are pouring hundreds of billions of dollars into AI infrastructure. However, Alibaba Chairman Joe Tsai has warned that this aggressive buildout may be forming a bubble, as many data centres are being constructed without confirmed demand or clear customers. Speaking at the HSBC Global Investment Summit, Tsai criticised the speculative nature of these projects and questioned the necessity of such outsized spending, arguing that current investments are far outpacing actual demand for AI services.

Ironwood: The first Google TPU for the age of inference
Google introduced Ironwood, a 7th-generation Tensor Processing Unit (TPU) and the company’s first designed specifically for inference. Unveiled at Google Cloud Next 25, Ironwood is Google’s most powerful and energy-efficient TPU to date, with each chip delivering a peak 4,614 TFLOPs of compute and offering twice the performance-per-watt of its predecessor, Trillium. Systems built on Ironwood can scale up to 9,216 liquid-cooled chips per pod, achieving 42.5 exaflops of compute—over 24 times the power of the world’s largest supercomputer, El Capitan, which offers 1.7 exaflops per pod. Ironwood will be available to Google Cloud customers later this year.

Nvidia Blackwell Ahead in AI Inference, AMD Second
Nvidia’s new Blackwell B200 GPU topped the latest MLPerf benchmarks, outperforming all competitors in large language model tasks, including new tests for real-time and high-context AI. AMD’s Instinct MI325X closely trailed Nvidia’s H200, showing strong performance gains from added memory. Intel’s new Xeon 6 CPUs improved over past versions but lagged behind GPU-based systems, and Google’s TPU v6e showed moderate progress in image generation, delivering performance roughly on par with Nvidia’s H100.

AI masters Minecraft: DeepMind program finds diamonds without being taught
Google DeepMind’s new AI system, Dreamer, has successfully learned to collect diamonds in Minecraft without human guidance—a major milestone in the game, as anyone who has ever played Minecraft will know. Unlike previous systems that relied on human gameplay data, Dreamer uses reinforcement learning and builds an internal "world model" to imagine and plan actions, allowing it to adapt to unfamiliar environments and generalise knowledge across tasks, a key step towards more flexible, general-purpose AI.

DeepCoder: A Fully Open-Source 14B Coder at O3-mini Level
Together AI and Agentica Project present DeepCoder—a fully open-source code reasoning model fine-tuned from Deepseek-R1-Distilled-Qwen-14B. According to benchmarks provided by Together AI, DeepCoder matches o3-mini performance with just 14B parameters.

Inside Google’s Two-Year Frenzy to Catch Up With OpenAI
Google was the undisputed leader in AI—until OpenAI released ChatGPT and turned the tech world upside down. This article documents how Google went from initial shock and panic to launching a frantic, high-stakes sprint to catch up. Over two turbulent years, the company rushed out products, merged its AI labs, released the Gemini family of models, and embedded generative AI across its products, all while grappling with internal chaos, ethical missteps, staff burnout, and political backlash.

Mira Murati’s AI startup is reportedly aiming for a massive $2B seed round
Thinking Machines Lab, the new AI startup founded by former OpenAI CTO Mira Murati, is reportedly seeking a staggering $2 billion in seed funding, which could be the largest seed round in history, potentially valuing the company at over $10 billion despite having no product or revenue yet.

First Therapy Chatbot Trial Yields Mental Health Benefits
Dartmouth researchers have completed the first clinical trial of an AI-powered therapy chatbot. The chatbot, named Therabot, showed significant reductions in symptoms of depression (51%), anxiety (31%), and eating disorders (19%) among 106 participants. The study found that users formed strong therapeutic bonds with the chatbot, often engaging with it like a trusted friend. While promising as a tool to expand access to mental health support, researchers emphasised the need for clinician oversight and strict safety protocols due to potential risks in AI-driven mental health care.

Expert used ChatGPT-4o to create a replica of his passport in just 5 minutes bypassing KYC
One security expert found that Image Generation, ChatGPT’s latest feature responsible for the wave of Ghiblified images of everything, is capable of creating convincing fake passports realistic enough to bypass automated Know Your Customer (KYC) checks. “The implications are obvious—any verification flow relying on images as ‘proof’ is now officially obsolete. The same applies to selfies. Static or video, it doesn’t matter. GenAI can fake them too. Photo-based KYC is done. Game over,” he writes in a tweet.

---

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

---

🤖 Robotics

Kawasaki’s wolf-inspired, four-legged robot lets riders traverse uneven terrain

Robot dogs are cool, but have you ever wanted to ride one? If so, then Kawasaki has something for you. Meet CORLEO—a hydrogen-powered, four-legged robot designed to be ridden like a futuristic, all-terrain motorcycle. Inspired by wolves and agile animals, CORLEO features a hydrogen fuel cell, 150cc engine, and AI systems for autonomous leg control and navigation. Although this is only a concept and CORLEO exists only in renders and as a mock-up (Kawasaki projects launching CORLEO within the next 25 years), it presents a bold vision for the future of mobility. Also, it just looks cool, and I’d love to ride one someday.

Waymo readies autonomous cars for first international tests in Japan
Waymo is set to begin its first international test of driverless vehicles in Japan next week. The test will involve 25 manually driven cars collecting mapping and traffic data across seven central Tokyo wards to study Japan’s left-hand traffic and dense urban conditions. While there’s no confirmation of a future commercial robotaxi launch, Waymo calls the move a “significant milestone.”

Samsung’s Ballie robot launches in the US ‘this Summer’ with Gemini in tow

Samsung has announced that its long-awaited Ballie home robot will officially launch in the US and Korea in summer 2025. Powered by Google’s Gemini and Samsung’s own language models, Ballie will use multimodal inputs—including voice, visuals, and environmental data—to provide real-time assistance, such as outfit recommendations and wellness tips. Initially revealed in 2020, the redesigned robot also includes features like a built-in projector, music playback, controlling smart devices, call handling, and pet monitoring.

▶️ Autonomous Human-Robot Interaction via Operator Imitation (3:04)

In this video, featuring one of those adorable Star Wars real-life droids, researchers at Disney show how they trained a robot to interact with people autonomously by imitating expert operators. Using a transformer model, the robot learns both motion and mood from real human-robot interactions. The result is a system that can perform natural, recognisable behaviours—no remote control needed—even on different robot platforms.

Humanoid Robots Are Lousy Co-Workers. China Wants to Be First to Change That.
Chinese startups are rapidly advancing humanoid robot technology as the country aims to lead the global market by 2027, writes The Wall Street Journal in this article. Companies like UBTech and Unitree are deploying humanoid robots in factories and warehouses to learn on the job and improve efficiency. While still clumsy and expensive, these humanoids are expected to evolve quickly, driven by China’s manufacturing scale and state support. Meanwhile, US companies developing similar technologies face challenges in keeping pace with China’s integration of robots into real-world environments.

Robotics Meets Runway: Unitree G1’s Catwalk Debut at SHFW
Unitree Robotics’ humanoid robot, Unitree G1, made its debut appearance as a fashion model at Shanghai Fashion Week, walking the runway alongside human models in a fusion of fashion and technology. The event also featured Unitree’s quadruped robot and a collaborative cyber horse art installation with artist Yiming.

▶️ Autonomous Racing Drones: no map, no IMU, no SLAM (0:32)

These autonomous racing drones fly using only cameras—no prepared in advance maps and no other sensors, just cameras. It is an impressive feat of engineering to make these small drones fly through obstacles without colliding with anything. Bonus points for achieving this on a poorly lit theater stage.

🧬 Biotechnology

Miniaturized CRISPR Packs a Mighty Gene Editing Punch
NanoCas is a miniature gene-editing system that could overcome a major limitation of traditional CRISPR-Cas9: its large size. By shrinking the Cas9 protein to about one-third its original size, NanoCas fits into a single viral delivery system, allowing it to edit genes directly inside hard-to-reach tissues like muscle and potentially the brain. In tests on mice and monkeys, NanoCas successfully edited genes linked to high cholesterol and muscular dystrophy with promising efficiency and minimal side effects.

💡Tangents

The 'World's Most Advanced Microchip' Has Been Unveiled
TSMC unveiled the world’s most advanced 2-nanometre microchip, which promises major gains in speed, energy efficiency, and transistor density over its 3nm predecessor. The new chips are expected to enable smaller, lighter, longer-lasting consumer devices, as well as significant improvements in performance and efficiency for data centre and AI chips. Mass production is slated for later this year.

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!"