In partnership with
In Today’s Issue:
🚀 Anthropic’s annualized revenue skyrockets from $19 billion to $30 billion in a single month
🧮 OpenAI’s GPT-5.4 cracks a decades-old math problem
🪖 Ukraine dramatically scales its uncrewed ground vehicles
🧬 A Boston startup injects the first-ever cellular reprogramming therapy into human eyes
✨ And more AI goodness…
Dear Readers,
Meta just mass-shipped a frontier-class AI model to over a billion people, and the wildest part is that it might not even be today's biggest story. Muse Spark, built from scratch under Alexandr Wang, marks Meta's sharpest pivot yet, from open-source darling to proprietary powerhouse, proving that distribution can trump raw benchmark dominance.
But while Meta rewrites its AI playbook, Anthropic is rewriting the revenue charts entirely, rocketing from $19 billion to $30 billion in annualized revenue in a single month, with whispers of $100 billion on the horizon and a $10 billion training bill for "Mythos." Meanwhile, OpenAI's GPT-5.4 is solving decades-old math problems that stumped human researchers, Ukraine's robot army just completed 21,500 ground missions in one quarter, and a tiny Boston startup is injecting the first-ever cell rejuvenation therapy into human eyes, attempting to literally reverse aging organ by organ. Today's issue covers a world where AI is simultaneously cracking pure mathematics, reshaping battlefields, fueling a multi-billion dollar arms race between labs, and now, potentially turning back the biological clock, so grab your coffee and let's get into it.
All the best,
Kim Isenberg
🚀 GPT-5.4 Cracks Another Erdős problems
OpenAI researchers used an internal AI model to solve five challenging Erdős problems, showing real progress in reasoning. One highlight: a long-standing question about graph coloring turned out to be false, the AI built a clever counterexample that breaks the expected rule in a simple, surprising way. The result shows AI isn’t just crunching numbers, it’s finding creative insights that even mathematicians enjoy unpacking.
Finding new solutions and ways that people haven't thought of before is very reminiscent of AlphaGo's legendary Move 37.
🤖 Ukraine Rapidly Expands Ground Robots
Ukraine is accelerating its use of uncrewed ground vehicles (UGVs), completing over 21,500 missions in Q1 2026, including 9,000 in March alone, a threefold jump since November. These robots are increasingly replacing soldiers in high-risk roles like supply delivery, mine-clearing, and casualty evacuation, while the number of military units using them surged from 67 to 167 in just months.
It shows how automation is reshaping warfare as manpower shortages and battlefield dangers intensify, though UGV deployment still trails far behind aerial drone operations in scale. With the advent of increasingly more robots, it was only a matter of time before they were widely used in wars.
🚀 Anthropic’s Explosive Revenue Growth Surge
Anthropic is scaling at breakneck speed, jumping from $19B to $30B annualized revenue in just one month and potentially on track to hit $100B within months if momentum holds. This surge may already place it ahead of competitors like OpenAI, while also accelerating its path to profitability and a possible IPO as early as late 2026.
However, sustaining this trajectory faces real constraints, massive compute demands (multi-gigawatt infrastructure), rising inference costs, and natural slowdown from market saturation. Even so, Anthropic’s growth is reshaping the AI race, forcing rivals and partners like Amazon Web Services and Google to rethink strategy, pricing, and scale. Oh, and by the way: Training "Mythos" reportedly cost $10b.
Claude Opus was used to map Anthropic’s Mythos onto the Epoch AI Intelligence Index (ECI), showing it tracks almost exactly on trend, slightly above GPT-5.4, but without accelerating the overall capability curve.
Demis Hassabis (CEO Google Deepmind): Why AGI is Bigger than the Industrial Revolution & Where Are The Bottlenecks in AI
The CRM That Actually Does the Work? Lightfield.
Most CRM failures start with a lack of context. They start as manual systems of record that just sit there, forcing you to do all the work. But holy moly: Lightfield just launched Skills and Knowledge, and I’ve realized that while the way we manually manage work doesn't scale, our judgment finally can.
The problem I've seen for so many founders is that your "secret sauce"—the specific questions that uncover a customer's real problems and the exact reasons people choose you—is completely trapped in your head. When you finally hire someone to help, they start from zero while your money and time run out.
Lightfield solves this by capturing every email, meeting, and transcript automatically to build a "centralized memory" of your business. This is their real superpower: unlike other tools that just give you a pile of transcripts, Lightfield stitches every conversation into a continuous understanding of the relationship so you never lose a key detail again. Then, it puts that context to work:
Knowledge encodes your specific business context—things like your ICP, pricing, and why you actually win.
Skills are reusable templates you can invoke for your workflows, like how you qualify deals or build proposals based on what was actually discussed.
The Result: You can prompt these Skills to run against your live data, ensuring your next hire starts from your best judgment, not from scratch.
We’ve started using it because time zones wreck us, plus it flags lost threads and drafts follow-up emails grounded in real conversation history. It even solves the "shadow decision-maker" problem by mapping the full buying committee from your communication history, identifying stakeholders who control the deal even if they’ve never been in a meeting.
It is a high-leverage experience that syncs with your calendar and inbox to ensure your best moves never stay trapped in your head again.
Meta's AI Comeback Begins
The Takeaway
👉 Muse Spark is Meta's first model built entirely from scratch under Alexandr Wang, scoring in the top five globally and dominating health and visual understanding benchmarks while trailing in coding and agentic tasks.
👉 The shift from open source Llama to proprietary Muse shows a fundamental strategy change: Meta is prioritizing product integration and potential API revenue over community goodwill.
👉 With over 10x compute efficiency gains compared to Llama 4 Maverick, Meta has proven it can build competitive models at a fraction of the training cost, a crucial advantage as infrastructure spending balloons toward $135 billion this year.
👉 The real competitive moat is distribution: Muse Spark ships free to billions of users across WhatsApp, Instagram, Facebook, and AI glasses, giving Meta a testing and feedback loop no other lab can match.
Nine months, one complete rebuild, and $14.3 billion later, Meta just dropped its most competitive AI model to date. Muse Spark, the first model out of Meta Superintelligence Labs under Alexandr Wang, is a clean break from the Llama era. Internally codenamed "Avocado," the model was built from scratch with new infrastructure, new architecture, and new data pipelines. The results speak for themselves: Muse Spark scores 52 on the Artificial Analysis Intelligence Index, placing it in the top five behind GPT-5.4, Gemini 3.1 Pro, and Claude Opus 4.6.
It dominates health benchmarks, beating all rival models on HealthBench Hard with a score of 42.8%, and excels in visual understanding. But Meta is honest about the gaps: coding and long-horizon agentic tasks remain weak spots, with Terminal-Bench scores trailing GPT-5.4 significantly.
What makes this interesting is the efficiency play. Meta achieved these capabilities using over an order of magnitude less compute than Llama 4 Maverick. The model is free, rolling out across WhatsApp, Instagram, and Meta's AI glasses, with a private API preview for select partners.
One thing must always be kept in mind: Meta doesn't need the best AI model. It only needs to be good enough to satisfy the majority of its over one billion users. The target group is presumably users who use ChatGPT-free in their daily lives. That's the benchmark that Meta's own model has to meet. And they've achieved that goal. Because their moat is distribution.
Why it matters: Muse Spark signals that Meta is serious about competing at the frontier after the Llama 4 stumble, backed by a rebuilt AI stack and a massive infrastructure bet of up to $135 billion in 2026. It is a very good model, although not quite on par with the best models like GPT-5.4 or Opus 4.6, but good enough for users in the Meta universe.
Sources:
🔗 https://ai.meta.com/blog/introducing-muse-spark-msl/
🔗 https://fortune.com/2026/04/08/meta-unveils-muse-spark-mark-zuckerberg-ai-push/
Meta has put a lot of effort into making their model one of the safest in the world. Not bad, Meta.
Cell Reset Therapy Enters Human Trials
The idea of making old cells young again just moved from the lab to the clinic. Boston-based Life Biosciences has raised $80 million in Series D funding to advance ER-100, the first partial epigenetic reprogramming therapy ever cleared by the FDA for human trials. Co-founded by Harvard geneticist David Sinclair, the company uses three Yamanaka factors to partially rewind a cell's biological clock, restoring youthful function without fully resetting the cell's identity.
(Jerry McLaughlin is CEO of Life Biosciences)
The first target: vision loss from glaucoma and optic nerve damage, where damaged neurons currently have no way to regenerate. The therapy is delivered via a single injection into the eye, then activated for eight weeks using the common antibiotic doxycycline as an on/off switch.
(David A. Sinclair)
What makes this so compelling is scale: The $80 million extends the company's runway into late 2027 and funds expansion into liver and metabolic diseases, while competitors like Altos Labs and Retro Biosciences remain preclinical. Life Biosciences is the first company to reach human trials with this approach, operating with a team of fewer than 20 people. If ER-100 proves safe and effective, it won't just be a new eye treatment. It will be proof that cellular aging can be reversed in humans, organ by organ!
Your first HR system, implemented right
Rolling out your first HR tool? Get a step-by-step guide to avoid common mistakes, drive adoption, and build a scalable HR foundation.