Thursday, July 30, 2026

PreciGenetics, Fractals, And A Whole New Space

To understand what PreciGenetics is up to, use fractals. It is about to open up a whole new space. And I have compared it to SpaceX. Too big for VC. As in, it needs to go to its mega customers.

And there are people who call me Param. :_

Musk’s Management

Silicon Valley: You Cannot Compile A Cell. When it comes to biology, AI advances started an existential stop-clock. We are about to either live in a world that gets much safer, or much more unsafe; there is no middle way. Silicon Valley realized the problem this summer, but it underestimates it; underneath almost everything written about AI and biosecurity sits one assumption, that biology will behave like software. This essay exists to express something it does not understand the gravity of: you cannot compile a cell. You cannot make a cell compile, no matter how much money, hype, employees and GPUs you use. Until Silicon Valley absorbs that, it’s building the future on a false premise. .............. yes, you can design a drug in milliseconds with some electricity, GPUs and an algorithm; then, you get to spend weeks to 7 years validating it. Then, it takes 2 billion dollars at a 97% failure rate and a 15-year-long process to validate reality in a human being; you hope no one dies in the process. .......... software and biotech are very different worlds; most of all, in their efficiency. When I first absorbed reality, I was either going to choose not to play the biotech game, or flip the game on its head. I could only live with one of those answers, it was not the former one; so, I chose the only way that is guaranteed to change this game. Build the hardware that makes a machine learning biological truth cost less and mean more in the human body. I founded Precigenetics, aiming to give the world the measurement infrastructure for intelligent biological discovery..........

For most of the history of biology, measuring a cell meant killing it.

........... when someone tells me “AI will solve biology” - something I hear a lot more in Silicon Valley than in Boston - I smile, and I ask: with what data? An LLM can hallucinate a plausible molecular structure in milliseconds. Knowing whether that molecule behaves predictably in a living system is the bottleneck, and the bottleneck remains physical. ......... A seatbelt is fine, but they do not help much if the driver is blind. Guardrails don't stop a determined adversary, they price out the amateur ones, and they hurt everyone else. We are handing out badly-designed seatbelts that handicap scientists trying to cure disease without actually saving the world from its gravest threats. If a bad actor does not care who the car crashes into, it can drive drunk and blind. Open-weights models exist, and they keep improving. No one in America can control what drugs are designed and manufactured in China without massive consensus; this is not a one-party issue. Anyone can run a capable drug design model on their own hardware without asking anyone’s permission. This is not a policy failure awaiting better policy; it is technological failure, so it must be awaiting better technology. .............. There is Boston, and Silicon Valley. Then, there is Washington. Here is why I needed to, not wanted to, start Tycho: I could not sleep at night anymore. I thought biological validation is the big bottleneck in pharma - you don’t know what works - which it is, but then my eyes opened to the most difficult truth I have ever come around: what if you do not care what helps a human? An attacker doesn’t need a molecule that would survive clinical trials. They don’t need FDA-grade validation or a clean dose-response curve. They need it to work once; if it hurts someone, that's what they want, and if it does not, it stays silent. The defender needs the defense to work every time, everywhere, forever. “Works once” versus “works always” is not a fair fight, and it’s not a fight you win by negotiating with Fable. The only way to defend is by being one dimension ahead of an adversary - having the ability to search, and rescue, from any molecule, at all times. That intelligence will exist one day, that is a fact. Whoever has it has biological power; for freedom and humanity to exist, it needs to be America. .............. The model was the bottleneck. The physical world is the bottleneck; for the attacker and for Washington. Which means the physical world is where the defense has to live. ............. The software industry has already been through this exact doctrinal evolution. It happened in cybersecurity, so biologists weren’t watching. Early infosec believed in secrecy, which meant: hide how the operating system works and keep the vulnerabilities quiet; control who knows what. It failed the way it was always going to fail: knowledge can’t be gate kept forever, so adversaries share notes, and the defenders ended up as the only people who didn’t understand the system. .......... Eventually, the field grew up. It adopted “assume breach,” which plainly stands for “stop pretending you can prevent information from spreading.” Get one dimension ahead of the threat with firewalls, endpoint detection and threat intelligence. Instead of stopping unstoppable movement, instrument the system so that when something hostile moves through it, you see it move, and you respond. We gave up on prevention of knowledge and chose something far more intelligent: detection of events. That one shift is the reason you can do your banking online. Now ask the hard question: what does endpoint detection mean when the endpoint is a cell? ............... It means a measurement layer. Instruments that read living biology, continuously, in the physical world, so that anomalous biology is something we observe as it begins, not something we reconstruct from samples after the fact. In biology, detection infrastructure is measurement infrastructure. There is NO imaginable software-only version of this. You cannot firewall a cell; you can only learn to see it, and then learn to change the cell's state, across time, for which your measurement must keep it alive. That is what we’re building. ............. Four hundred years ago, before the telescope, a Danish astronomer named Tycho Brahe decided the problem with astronomy was that nobody could actually see the sky properly. He spent decades building the most precise observation instruments of his era and cataloguing the positions of the stars and planets, night after night, with a fanaticism his contemporaries found excessive. He never got the theory himself. He didn’t have to. A young assistant named Kepler took Tycho’s tables and derived the laws of planetary motion from them. The catalogue came first; the laws came from the catalogue.

Measurement infrastructure precedes theory in every single field, every single time.

Biology has spent a century trying to skip that step. We named our new real-time biosensing effort Tycho. The name is self-explanatory and complete, and it will never change. ................... Today, if something anomalous starts happening in biology anywhere on earth, the default state of the world is that nobody sees it until the consequences walk into an emergency room. .............. you cannot defend against what you cannot see, and biology has never had the instruments to see in real time. Now it can. ..............

Cleopatra

, a toxicity foundation model, built from infrastructural leaps similar to Tycho. The ambition is an AlphaFold-like moment for toxicity: a model that learns what molecules do to living human cells, trained on continuous single-cell measurements that record which chemicals are inside each cell and how their concentrations change over time, so characterizing a compound’s toxicity takes hours instead of months. ...........

Cleopatra and Tycho are twins; both say: pair intelligence with sensing and you have biology’s endpoint detection and threat intel. Same infrastructure, two futures: one about curing disease faster, one about catching harm earlier. This is dual-use in the best direction.

............. Nobody made electricity safe by deciding who was allowed to generate it. They made it safe with insulation and circuit breakers: protection and detection built into the physical layer, where the danger actually lives. Every restriction aimed at the hypothetical bad actor lands, with perfect reliability, on the actual good ones: the academic labs, the startups, the people designing cures for the worst diseases humanity faces everyday, in kids and adults and even other organisms. Meanwhile, look at what is happening: the dedicated adversary will always route around your closed approach in a weekend. The bottom line is: stifling open progress disarms the defenders and inconveniences literally no one else.

at No comments:
Labels: , ,
Subscribe to: Posts (Atom)