Weekly Dose of Optimism #132

Curing Blindness, Genetic Womb Treatment, Evo 2, AI co-scientist, Topological Qubits, Robots

Hi friends 👋,

Happy Friday and welcome back to our 132nd Weekly Dose of Optimism. This isn’t a Chocolate Factory, but all this Gene stuff is about to get a lot Wilder. If that makes zero sense to you now, keep reading.¹

Let’s get to it.

---

Today’s Weekly Dose is brought to you by… Cache

Is your financial future riding on just one or two stocks?

Are you incurring huge tax bills when you sell your appreciated stocks, or holding too much in one or two stocks? The Cache Exchange Fund, benchmarked to the Nasdaq-100, is built to diversify your stocks while deferring taxes. Everyone from engineers to CEOs have turned to Cache to manage their concentrated stocks.

New investors are onboarded bi-weekly, with the next fund closing on Feb 28th.

Learn more about the Cache Exchange Fund.

---

(1) Functionally blind man gets sight back after gene therapy

Fergal Bowers for RTE via Irish Friend of the Dose Will O’Brien

A 31-year-old Sligo man who was functionally blind has got his sight back, after being treated with a new gene therapy at the Mater University Hospital in Dublin. He is the first patient to receive the ground-breaking ocular gene therapy treatment 'Luxturna' in Ireland.

Would you look at that!

You know the world is becoming a better place when you have to use like the fifth best joke about a blind person regaining sight to intro a story because you’ve used the other top four jokes already to cover other stories about a blind person regaining sight.

A functionally blind Irish man regained much of his vision after receiving Luxturna, a groundbreaking ocular gene therapy, at Mater University Hospital in Dublin. He had previously suffered from a rare inherited retinal dystrophy for over a decade but experienced significant improvement within weeks of treatment. The procedure involves injecting a modified virus carrying a functional copy of the faulty gene into the retina (which doesn’t sound super pleasant tbh), enabling cells to produce the missing enzyme needed for vision. Professor David Keegan, the lead surgeon, called this a major step in precision medicine, with a national registry now being developed to identify more patients.

I can’t wait to see what happens next.

(2) Rare genetic disorder treated in womb for the first time

Smriti Mallapaty for Nature

A two-and-a-half-year-old girl shows no signs of a rare genetic disorder, after becoming the first person to be treated for the motor-neuron condition while in the womb. The child’s mother took the gene-targeting drug during late pregnancy, and the child continues to take it.

More cool gene therapy stuff.

For the first time, a fetus was successfully treated for spinal muscular atrophy (SMA), a severe genetic disorder that typically leads to fatal motor neuron degeneration in infancy. The child's mother took Risdiplam, a gene-modifying drug, during the final six weeks of pregnancy, and the child has continued treatment post-birth. Now nearly three years old, the child shows no signs of the disease, which is unprecedented for cases this severe. Hell yeah!

This breakthrough suggests that early, in-utero intervention could significantly improve outcomes for SMA, which remains a leading genetic cause of infant mortality. The idea originated from the parents (talk about being good parents!), who had lost a previous child to the disease, and the FDA approved the treatment for this one case (go…FDA?). This is a one-of-one case, but if replicated, this approach could shift how we treat diseases before birth.

(3) Genome modeling and design across all domains of life with Evo 2

From Arc Institute

Evo 2 learns from DNA sequence alone to accurately predict the functional impacts of genetic variation—from noncoding pathogenic mutations to clinically significant BRCA1 variants—without task-specific finetuning. Applying mechanistic interpretability analyses, we reveal that Evo 2 autonomously learns a breadth of biological features, including exon–intron boundaries, transcription factor binding sites, protein structural elements, and prophage genomic regions.

If you think curing blindness and rare genetic disorders in the womb with genetic therapy is crazy, then oh baby, you better strap the fuck in, because the next decade is going to be wild.

Evo 2 is a new AI model from a team at the Arc Institute that learns the “language” of DNA, allowing it to predict the effect of mutations, design new sequences, and influence gene expression.

The model is trained on 9.3 trillion DNA base pairs across all domains of life. It predicts whether mutations are harmful or benign without specific training on human disease data, outperforming specialized models and accurately handling noncoding mutations—suggesting it has internalized DNA’s fundamental principles. That’s right: Evo 2 has learned the core rules of DNA on its own, allowing it to predict genetic effects across all life forms without being specifically trained on human disease data.

Evo 2 can create entire DNA sequences, like those for mitochondria and bacteria, in a way that looks natural. It can also control how genes are turned on or off by influencing DNA structure. As a test, researchers used it to encode Morse code into DNA, showing its potential for designing custom genetic systems.

One more time for those in the back: like God and/or evolution, Evo 2 can create entire DNA sequences.

Better yet, Evo 2 is fully open-source, making genome design more accessible and increasing the pace of innovation in bioengineering. In the future, we won’t just study biology, we’ll code it.

(4) Accelerating scientific breakthroughs with an AI co-scientist

Juraj Gottweis, Google Fellow, and Vivek Natarajan, Research Lead

We introduce AI co-scientist, a multi-agent AI system built with Gemini 2.0 as a virtual scientific collaborator to help scientists generate novel hypotheses and research proposals, and to accelerate the clock speed of scientific and biomedical discoveries.

If it wasn’t yet abundantly clear already, the pace of scientific research and discovery is about to absolutely take off. To that end, this week Google introduced AI co-scientist, a multi-agent system designed to assist researchers by generating novel hypotheses, research plans, and experimental protocols.

This isn’t just a powerful LLM that churns through existing research and summarizes the literature. AI co-scientist mimics the scientific method, using specialized agents to iteratively refine ideas and improve output quality, allowing it to generate novel research ideas. It operates like a scientist: self-critique, ranking, feedback loops. And not just any scientist, a successful scientist.

So far, early testing has shown promising results in biomedical research, things like drug repurposing and disease targeting discovery. It has also demonstrated the ability to rediscover novel mechanisms in antimicrobial resistance, validating its potential as a scientific collaborator. Meaning, AI co-scientist came to a similar conclusion as human researchers, but without exposure to their unpublished work. It scientific method-ed its way to the same conclusion.

If AGI looks something like automated expert level and breakthrough research, then boy, AGI feels closer and closer. What a great time to be a human.

(5) Interferometric single-shot parity measurement in InAs–Al hybrid devices

From Microsoft Azure Quantum (explain in NYT here)

We implement a single-shot interferometric measurement of fermion parity in indium arsenide–aluminium heterostructures with a gate-defined superconducting nanowire…The large capacitance shift and long poisoning time enable a parity measurement with an assignment error probability of 1%.

The four states of primary matter: solid, liquid, gas, and topological qubit.

That’s according to new research out of Microsoft, which claims to have developed this new physical state that can be harnessed to power quantum computing. A topological qubit is a special kind of qubit that stores information in the way certain particles move and interact, making it much more stable and resistant to errors than regular qubits. There’s that error reduction bit again. Any time you read about advancements in quantum computing, it generally is about some novel way of error reduction.

So what exactly did Microsoft do? The team developed a device that can measure the state of special particles called Majorana Zero Modes (MZMs). MZMs are special quantum particles that act as their own antimatter and can store quantum information in a way that makes them resistant to errors. Using a superconducting nanowire and quantum capacitance measurements, they detected these states quickly and with high accuracy—making errors only 1% of the time. The measured states remained stable for over a millisecond, long enough to be useful for computation.

Anyway you cut it, this research is a pretty big advancement on the road towards useful quantum computing. That said, there is still some skepticism around whether the observed signals truly come from MZMs in a topological phase or from more conventional quantum states that mimic their behavior. Without that definitive proof that these are genuine MZMs, it’s still a breakthrough but the path towards scalable useful quantum is way less clear.

Microsoft CEO Satya Nadella went on Dwarkesh to talk about his company’s quantum breakthrough, AI video game creation, and path to smarter and smarter machines.

The whole thing is worth a watch, but the part we agree with most is Satya’s assertion that you shouldn’t believe AGI is here when any lab tells you it is, but when GDP starts growing at 10% a year. Again - just a really great time to be a human.

(6) Big Week for the Robots, Too

Packy here. Not only are we adding a sixth story, but it’s a two-for-one. Massive week. We didn’t even mention Grok 3, which is very cool but not cool enough to make the cut. But we would be remiss if we didn’t mention the robots.

On Thursday, the Polish company Clone released a video of its very human / Westworld looking Protoclone hanging in midair and twitching to a dark cinematic track by Ludwig Göransson, which was creepy but still pretty amazing, especially given that PitchBook says Clone has only raised $7.1 million, which at least we can agree to put to rest once and for all the joke that Polish people are stupid, right?

On the other end of the funding spectrum, Figure, which is rumored to be raising $1.5 billion at a $38 billion pre-money valuation (I would not be described as the most valuation sensitive investor in the world and I think Tech is Going to Get Much Bigger than people expect, including humanoid robots eventually, but that valuation beggars even my formidable belief) released a mindblowing video of two of its Figures running on its new Helix Vision-Language-Action model.

Once again, a bit creepy watching them communicate without verbally communicating (robots talking to each other would be skeuomorphic and inefficient, for our comfort only, but its absence is certainly discomforting). BUT watching the one Figure hand the bag of cheese to the other Figure was one of those moments that makes you realize “Oh, shit, the future is going to be even more different than I expected because while it makes total sense, I’d just never pictured multiple robots working with each other.”

All of this — the AI breakthroughs, the robots — can seem a little bit scary from the perspective of a human. But personally, I hate diseases and I don’t particularly like putting away my groceries, and I for one welcome our robot underlings.

BONUS: Packy on Sourcery with Molly O’Shea

Packy went on Molly O’Shea’s podcast Sourcery last week. They covered a bunch — scaling the newsletter, his approach to writing, and the types of companies that most interest him today. Give the pod a listen, or if Packy is not your cup of tea, then at least check out Molly’s work and give her a follow.

---

Have a great weekend y’all.

Thanks to Cache for sponsoring! We’ll be back in your inbox next week.

Thanks for reading,

Packy + Dan

1. PM Writes: This makes zero sense to me.