What this covers

Jacob Kimmel argues with Dwarkesh Patel that aging is uniquely tractable because evolution never optimized against it. The core logic rests on three evolutionary constraints: high ancestral hazard rates from predation, accident, and infection meant few humans lived long enough for longevity traits to be selected; kin-selection worked against aged, low-fitness individuals; and the genome's optimization budget—bounded by mutation rates and population size—was dominated by infectious-disease resilience, leaving lifespan as a neglected problem. This makes aging fundamentally different from challenges evolution worked hard to solve, and therefore more amenable to engineering intervention. Kimmel then traces the path forward: transcription factors serve as evolution's modular levers, where small edits to a few genes produce large phenotypic effects, making them ideal targets for reprogramming. The breakthrough example is Shinya Yamanaka's four-factor reprogramming of adult cells to embryonic state—proof that age and cell identity can be reset simultaneously.

The major technical obstacle is measurement and scale. Yamanaka's screen succeeded because success was obvious (cells turned blue) and reproducible cells amplified into detectable colonies; aging has neither quality—old and young cells look molecularly similar, and reprogrammed cells don't self-amplify. This forces researchers to search an intractable ~10^16 transcription factor combination space. Kimmel's proposed solution is the virtual cell: a predictive model trained on sparse single-cell perturbation data that can forecast the effects of untested gene combinations in silico, bringing machine-learning-like scaling to drug discovery. The conversation also covers delivery mechanisms (why lipid nanoparticles and viral vectors are complementary), the economics of longevity medicines (how they could shift healthcare burden from expensive acute care to cheaper pharmaceutical systems), and why the pharmaceutical industry has failed to keep pace with Moore's Law—a phenomenon Kimmel calls Eroom's Law. Throughout, the discussion hinges on the meta-principle that problem selection, not execution, is what separates breakthrough science from incremental work.

Sharpest takeaway

Kimmel argues that aging is not a problem evolution ever optimized against—because high ancestral hazard rates, kin-selection regularization, and optimization constraints all weakened selection for longevity—which makes aging relatively tractable to engineer, and that epigenetic reprogramming via combinations of transcription factors, guided by predictive 'virtual cell' models, is the most promising path to add healthy decades to human life.

  • Evolution didn't optimize for longevity due to high baseline hazard rates, kin-selection against aged low-fitness individuals, and genome optimization constraints favoring infectious-disease resilience
  • Transcription factors are evolution's modular 'levers' on the genome, so small perturbations yield large phenotypic changes usable to reverse cell age
  • Predictive models trained on sparse single-cell perturbation data can search the intractable ~10^16 TF combination space to find age-reversing interventions

The claims · ranked40 claims · weighted by value

This asset isn't compiled yet

You're seeing its claims, ranked. Compile it to build the argument threads, weight them, and check each claim against your library — the full view.

0.86

Humans never evolved their own antibiotics because bacteria and fungi, with enormous population sizes and high tolerable mutation rates, are locked in a rapid Red Queen arms race; even if a human genome stumbled into making an antibiotic, pathogens would mutate around it quickly, so it is not a stable evolutionary investment for slow-mutating metazoans.

causalhigh valueestablishednovelty 3/4durability 4/4· Jacob Kimmel

“Even if our human genome stumbled into making an antibiotic, most pathogens probably would have mutated around it pretty quickly.”

0.85

The human gene TRIM5alpha once protected primate ancestors against an HIV-like pathogen but lost that capacity to restrict a different endogenous retrovirus; because that record persists in the sequence, making just a few edits (in the tens) to TRIM5alpha can restore dramatic restriction of HIV, demonstrating that historical lost defenses can be resurrected.

factualhigh valueestablishednovelty 4/4durability 4/4· Jacob Kimmel

“You can edit it back to actually restrict HIV dramatically.”

0.81

Yamanaka's reprogramming screen succeeded because of two features absent in aging research: success was trivially measurable (a single reporter gene turning cells blue) and successful cells amplified (proliferating into detectable colonies even at 0.001% efficiency); aging has neither—old and young cells look similar requiring complex molecular measurement, and reprogrammed cells don't amplify—so a vastly larger TF combination space (~10^16) must be searched, requiring predictive models.

causalhigh valueestablishednovelty 3/4durability 3/4· Jacob Kimmel

“There were two features of Shinya's problem that I think make it amenable to that sort of interrogation that aren't present for many other types of problems.”

0.76

Perturb-seq (introduced in three near-simultaneous 2016 papers) took a decade to bear fruit because early readouts were expensive and unreliable—only ~50% of cells revealed which perturbation was applied—and combinatorial perturbations compounded mislabeling (roughly 1/2^n correctly labeled for n genes); as sequencing cost dropped from dollars to fractions of a cent per cell and labeling improved, experiments scaled to millions of cells per day.

causalhigh valueestablishednovelty 3/4durability 2/4· Jacob Kimmel

“When you sequenced a cell, only about 50% of the time could you even tell which perturbation you put in.”

0.75

Large pharma operates like venture capital firms, externalizing early R&D to nimble biotech startups: roughly 70% of approved molecules originate in small biotechs even though most R&D dollars sit on big-pharma balance sheets, with pharma acting as an oligopsony of buyers for phase-one/phase-two-ready assets and running the expensive trials that account for much of the cost difference.

factualhigh valueestablishednovelty 2/4durability 3/4· Jacob Kimmel

“it's something like 70% of molecules approved in a given year come from originally small biotechs rather than large pharmas, even though you look at the actual dollars of R&D spend on the balance sheet and it's largely in big pharma.”

0.75

Eroom's Law (Moore's Law spelled backward) describes the consistent decrease since the 1950s in the number of new medicines invented per billion dollars invested in biopharma, persisting across many technological transitions—an inverse of compute's Moore's Law trajectory.

factualhigh valueestablishednovelty 2/4durability 3/4· Jacob Kimmel

“there's a very consistent decrease in the number of new molecular entities, new medicines that we're able to invent, per billion dollars invested.”

0.74

Most current drugs exert their effects by taking 'bank shots'—hitting receptors or signaling pathways upstream that ultimately change transcription factor activity—because TFs themselves are hard to drug directly: small molecules can enter the nucleus but are too small to disrupt the large TF-DNA binding interface, and recombinant proteins/antibodies are too large to cross the cell membrane.

causalhigh valueestablishednovelty 3/4durability 3/4· Jacob Kimmel

“many of our medicines, ultimately downstream, are leading to changes in TF activity, but we haven't been able to directly target them due to their physical location within cells.”

0.74

Almost every cell in the body is surveilled by the immune system except a few immune-privileged compartments (brain, eye, joints, ear), so engineered-cell delivery covers nearly the whole body; this makes cell-based and virus-based delivery complementary, since gene-therapy viruses are best suited precisely to those immune-privileged sites where immunogenic drugs are tolerated.

factualhigh valueestablishednovelty 3/4durability 3/4· Jacob Kimmel

“the shadow of all the diseases you can't address with viruses is what you can address with cells. Given the complementarity between them, you can probably cover the entire body.”

0.74

Breaking just one gene—the mitochondrial transcription factor TFAM—in one specific subset of T cells in mice dramatically shortens their lifespan, demonstrating that a single cell type can have outsized systemic effects and implying the inverse (rescuing one cell type benefiting the whole body) may also hold.

factualhigh valueestablishednovelty 3/4durability 3/4· Jacob Kimmel

“You can break a gene in there that encodes for a transcription factor in their mitochondria called TFAM, and you dramatically shorten the lifespan of mice.”

0.73

Because the baseline hazard rate during most of human and primate evolution was very high—integrating predation, accidents, and infection—few individuals lived long enough for longevity phenotypes to be selected, so very little gradient signal flowed back to the genome favoring extended lifespan.

causalhigh valuecontestednovelty 3/4durability 3/4· Jacob Kimmel

“From the best evidence we have, the baseline hazard rate was very, very high. Even absent aging, you're unlikely to actually reach those outer limits of possible health where aging is one of the main limitations.”

0.73

Long-durability medicines whose benefits accrue years later face a reimbursement problem because Americans change insurers every 3-4 years, so no insurer is incentivized to cover a benefit realized in the out-years; a pay-for-performance model—paying a fraction of the cost each year contingent on demonstrated ongoing benefit—aligns incentives, treating the therapeutic's cost-lowering effect like a pre-existing condition that follows the patient between payers.

normativehigh valuecontestednovelty 3/4durability 3/4· Jacob Kimmel

“the average person churns insurers every three to four years... no insurer is technically economically incentivized to cover that.”

0.73

Evolution's optimization is constrained: mutation rate bounds step size (too high causes cancer, too low prevents adaptation) and population size bounds parallel variant screening; with infectious disease being a dominant selective pressure, the weighted loss term placed far more weight on disease resilience than on longevity, so the genome never optimized for living forever.

causalhigh valuecontestednovelty 3/4durability 3/4· Jacob Kimmel

“One of the dominant sources of selection on the genome is really prevention of infectious disease.”

0.73

A 'virtual cell' model—trained on sparse single-cell perturbation data mapping which genes are turned on/off to resulting cell states—can leverage biology's high mutual information to predict in silico the effects of untested gene-perturbation combinations, enabling discovery of interventions that move diseased cells toward healthy states and giving drug discovery the compounding returns ML enjoys.

factualhigh valuecontestednovelty 3/4durability 3/4· Jacob Kimmel

“by doing some sparse sampling of those pairs... Once I've trained a model to predict from the perturbations to the output on the cell state, you can start to ask what would happen for some arbitrary combinations of genes.”

0.73

Because aging falls into the category of problems evolution did not optimize against, intervening to provide health should be relatively easy compared to problems evolution worked hard on; the fact that simplistic modern medicines—turning off a single gene everywhere at once—provide massive benefit is evidence supporting this.

causalhigh valuecontestednovelty 3/4durability 3/4· Jacob Kimmel

“it should be, relatively speaking, relative to a problem evolution had worked on, easy to try and intervene and provide health.”

0.73

The biggest risk in drug discovery is not figuring out how to make a molecule (e.g., an antibody) but figuring out what to target—which gene to intervene on to treat a disease; getting better at making molecules for gene X doesn't transfer to knowing what gene to target for disease Y, which is why expertise in drug discovery hasn't compounded across indications.

causalhigh valuecontestednovelty 3/4durability 3/4· Jacob Kimmel

“Figuring out what to make an antibody to target is the hard thing about drug discovery.”

0.73

To bring ML-like scaling economics to biotech, medicines must address large populations rather than the increasingly narrow genetically-defined diseases pharma has trended toward; since everyone ages and gets sick, age is one of the strongest predictors of disease across many conditions, so making cells younger could benefit nearly everyone (TAM approaching all of Earth), maximizing value per medicine.

normativehigh valuecontestednovelty 3/4durability 3/4· Jacob Kimmel

“All of us will one day get sick and die. So arguably, the TAM for any really successful medicine could be everybody on planet Earth.”

0.73

Because roughly a third of Medicare costs are spent in the final year of life and most healthcare expense concentrates once someone is already very sick, medicines that keep people healthier longer (incretin mimetics, reprogramming therapies) would shift burden from the expensive administration system to the more efficient pharmaceutical system—where technology makes care cheaper as drugs go generic—driving total healthcare spend down.

causalhigh valuecontestednovelty 3/4durability 3/4· Jacob Kimmel

“Something like a third of all Medicare costs are spent in the final year of life”

0.73

Gene duplication lets evolution solve new problems by copying a gene so the original retains its function while the duplicate accumulates mutations freely; this allows traversal of fitness valleys (where intermediate edits would otherwise be deleterious), explaining how new genes with relatively few edits can arise despite a low per-base-pair mutation rate.

causalhigh valueestablishednovelty 2/4durability 4/4· Jacob Kimmel

“Through duplication, you can create a scenario where those first two edits are totally tolerated. They have no effect on fitness. You've got your backup copy, it's doing its job.”

0.73

Shinya Yamanaka showed that turning on just four transcription factors out of ~20,000 genes can revert an adult cell all the way to an embryonic stem cell, proving cell type and age can be reprogrammed simultaneously—but doing so changes cell identity and can be pathological (e.g., causing teratomas), so the challenge is reverting age without changing type.

factualhigh valueestablishednovelty 2/4durability 4/4· Jacob Kimmel

“just by turning on these four genes, turn an adult cell all the way back into a young embryonic stem cell.”

0.73

HIV, a retrovirus that integrates into CD4+ T cell genomes, encodes the protein Tat which interacts with the master immune transcription factor NF-kB to drive its own transcription; in cells where upstream TF machinery is deactivated it goes latent, creating a latent reservoir that lets it persist and evade immune clearance even when all actively-replicating cells are killed.

factualhigh valueestablishednovelty 2/4durability 4/4· Jacob Kimmel

“HIV encodes a protein called Tat, and Tat actually activates NF-κB.”

0.72

Most of science is problem selection rather than execution—you don't get better at pipetting or running experiments after a certain age, but you do get better at picking what to do—which is what made Yamanaka's choice of an easily-measurable, amplifying problem so remarkable.

normativehigh valuecontestednovelty 2/4durability 4/4· Jacob Kimmel

“Most of science is problem selection. You don't actually get better at pipetting or running experiments after a certain age, but you do get better at picking what to do.”

0.72

Transcription factors function as evolution's levers on the genome because they let a very small number of base-pair edits produce very large phenotypic changes—breaking one TF can delete a whole cell type, retargeting one can change hundreds of downstream genes—and because mutations are inherently small, evolution requires this small-edit/large-effect substrate, which also makes TFs a favorable, gradient-friendly target for medicines.

causalhigh valuecontestednovelty 4/4durability 3/4· Jacob Kimmel

“having transcription factors allows a very small number of base pair edits at the substrate of the genome to lead to very large phenotypic differences.”

0.69

The delivery problem for genetic medicines can be reduced to getting nucleic acids into any desired cell type specifically, since nucleic acids can drug anything in the genome; current modalities are lipid nanoparticles (fat bubbles defaulting to liver, tunable via attached antibodies) and viral vectors like AAV (limited packaging size), and sequence engineering can only add NOT gates to narrow delivery, never expand it to cells not initially reached.

factualhigh valueestablishednovelty 3/4durability 2/4· Jacob Kimmel

“If you can get nucleic acid into a cell, you can drug pretty much anything in the genome effectively.”

0.69

Under a selfish-gene view, extending an individual's lifespan without eliminating their fitness decline imposes a 'regularization' cost: an aged, less-fit individual's net calorie contribution to the genome is lower than allowing them to die and be replaced by two young individuals, so the genome may optimize for turnover at maximum fitness rather than longevity.

causalhigh valuecontestednovelty 3/4durability 2/4· Jacob Kimmel

“a population being laden demographically with many aged individuals, even if they did have fecundity persisting out some period later in life, is actually net negative for the genome's proliferation and that really a genome should optimize for turnover and population size at max fitness.”

0.69

Unlike LLMs, which were built on internet data that was a freely shared common good usable across domains, high-quality biological perturbation data is still in its infancy with no internet-like equivalent; this forces companies like NewLimit to vertically integrate and generate proprietary data to build the best models, analogous to being in the early 1980s of the web.

factualhigh valuecontestednovelty 3/4durability 2/4· Jacob Kimmel

“unlike with LLMs, where a lot of the data that was necessary to build these was a common good... this biological data is still in its infancy.”

0.69

Medicines that preserve health (felt directly by consumers) rather than merely fix disease will shift toward a direct-to-consumer model, exemplified by Lilly's LillyDirect bypassing pharmacies, PBMs, and compounders; this consumer-demand-driven model lets long-term-benefit payments be abstracted away from the churning payer system via financing-over-time.

forecasthigh valuecontestednovelty 3/4durability 2/4· Jacob Kimmel

“over time we're going to move more toward a direct-to-consumer model for many of these medicines which preserve and promote health rather than just fixing disease.”

0.68

There is no single upstream 'bad gene' whose deactivation solves aging; aging results from layers of molecular regulation (notably the epigenome), so the most likely outcome of longevity medicine is adding multiple healthy years incrementally rather than a single magic pill that fixes everything at once.

factualhigh valuecontestednovelty 2/4durability 3/4· Jacob Kimmel

“I don't think that there is a single monocausal explanation for aging.”

0.66

The epigenome—the chemical marks determining which genes a cell uses—degrades and shifts with age, so cells fail to deploy the right genetic programs at the right times, reducing resilience to insults and increasing disease susceptibility; remodeling the epigenome back toward its youthful post-developmental state should address many diseases driven by cellular dysfunction.

causalhigh valuecontestednovelty 3/4durability 3/4· Jacob Kimmel

“the epigenome can degrade with age. It changes. The particular marks that tell your cells which genes to use can shift as you get older.”

0.66

Because the body is a highly interconnected system with endocrine organs (liver, adipose) signaling to other tissues, rescuing function in even one cell type or tissue produces broad knock-on health benefits; evidence includes young-liver transplant recipients showing reduced risk of multiple diseases and better overall survival, and bone-marrow (HSC) transplants curing unrelated diseases.

causalhigh valuecontestednovelty 3/4durability 3/4· Jacob Kimmel

“If you're able to rescue function even in one cell type in one tissue, you often have knock-on benefits in many places that you didn't initially anticipate.”

0.66

Skin sagging with age is likely not a cellular process but a failure of elastin polymerization: long polymerized elastin fibers form only during development, and afterward cells make the monomer units but fail to polymerize them into new fibers, so the eventual fix would require programming cells to an extra-physiological state capable of reinvigorating polymerization and repairing fibers.

factualhigh valuecontestednovelty 3/4durability 3/4· Jacob Kimmel

“you only polymerize it and form a long fiber during development. Then the rest of your life, you make the individual units of the polymer, but for reasons no one as far as I can tell understands, they fail to polymerize.”

0.62

Because epigenetic marks set during development persist for decades (your tongue doesn't turn into a kidney) and targeted edits have been shown to last 400+ cell divisions over years in vitro and at least a couple years in monkeys, reprogramming doses could in principle last a very long time; current data show effects lasting several weeks, suggesting monthly or few-monthly dosing rather than daily IVs.

forecasthigh valuecontestednovelty 3/4durability 2/4· Jacob Kimmel

“The existence proof that those epigenetic reprogramming events can last decades is that my tongue doesn't spontaneously turn into a kidney.”

0.61

Incretin mimetics like Ozempic (GLP-1/GIP-1) produce wide-ranging benefits—weight loss, cardiovascular disease, addictive behavior, possibly neurodegeneration—because they are hormones acting on a small number of receiving cells whose communication with the rest of the body propagates knock-on effects, illustrating that acting on few cells can yield body-wide health benefits.

causalhigh valuecontestednovelty 2/4durability 3/4· Jacob Kimmel

“just by acting on the small number of cells in your body which are receiving this signal, the interplay and the communication between those cells and the rest of your body seems to have many of these knock-on benefits.”

0.58

By around 2100, drug delivery will likely be solved the way evolution solved it—using engineered immune cells (T cells, B cells) that patrol the body, run AND-gate logic to sense specific environments, and release nucleic-acid payloads only when needed; these cells can be engrafted to persist for years and carry billions of base pairs of logic, overcoming the immunogenicity of viruses and the physical-size limits of LNPs.

forecasthigh valuespeaker onlynovelty 4/4durability 2/4· Jacob Kimmel

“the way we will be delivering these nucleic acid payloads is actually by engineering cells to do it”

0.58

The observation that great discoveries (e.g., in mathematics) cluster before age 30 is best explained not by sociological factors but by fluid intelligence being maximized around 25-30, the age at which population size during human evolution was maximal and thus where selection for fluid intelligence was strongest.

causalhigh valuespeaker onlynovelty 4/4durability 2/4· Jacob Kimmel

“A much simpler explanation is that for whatever reason, our fluid intelligence is roughly maximized at the time where the population size during human evolution was maximal.”

0.55

The genome's regulatory structure resembles the attention mechanism: transcription factors act like queries, the genome sequences they bind are like keys, and genes are like values—a structure that allows editing one embedding vector (a binding sequence) to produce dramatically different outputs, mirroring how attention recurs in neural structures.

definitionhigh valuespeaker onlynovelty 4/4durability 2/4· Jacob Kimmel

“TFs are like the queries. The genome sequences they bind to are like the keys. Genes are like the values.”

0.51

NewLimit's modeling structure parallels LLM training: a base model predicts how TF perturbations change every gene the cell expresses (analogous to pretraining a general world representation), built on protein foundation models, then separate prediction 'heads' impose value judgments like 'this transcriptome looks younger' (analogous to RL on a specific objective), and additional heads could target arbitrary desired cell states.

factualhigh valuespeaker onlynovelty 3/4durability 2/4· Jacob Kimmel

“you first have imitation learning with pre-training that builds a general-purpose representation of the world. Then you do RL about a particular objective... You are describing an extremely similar procedure”

0.48

Because processes like iPSC reprogramming never occur in nature, natural transcription factors aren't necessarily optimal for them; simple modifications—mutating one Yamanaka factor (as in Super-SOX from SOX2) or swapping domains between TFs—dramatically improve efficiency, suggesting the ideal future reprogramming medicines may be synthetic TFs that have never existed rather than compositions of natural ones.

forecasthigh valuespeaker onlynovelty 3/4durability 2/4· Jacob Kimmel

“iPSC reprogramming never happens in nature, so there's no reason to necessarily believe that the natural TFs are optimal.”

0.36

Despite gray-market peptides from China, traditional pharma will retain most revenue because US drug payment flows through payers rather than direct-to-consumer, so most patients will choose a low-co-pay prescription of the genuine drug over a sketchy vial, making the gray market a limited concern at scale.

forecastcontestednovelty 2/4durability 2/4· Jacob Kimmel

“If you have the opportunity to either order a sketchy vial off of some website from some company in Shenzhen, or you can go through your doctor and get a prescription with a relatively low co-pay for Tirzepatide, the real thing, most patients will go for Tirzepatide.”

0.35

Selecting for longevity is like a long-horizon reinforcement learning problem with a ~20-year horizon and a sparse scalar reward (number of surviving offspring), which is surprising it works at all given how difficult RL is for models even on intermediate hour-long goals.

factualspeaker onlynovelty 3/4durability 2/4· Dwarkesh Patel

“It's a long-horizon RL problem, a 20-year horizon length, and then there's a scalar value of how many kids you have”

0.30

The way we are coming to understand how the brain works is by first-principles engineering of intelligence in AI, after which each AI mechanism (like attention with positional encodings) turns out to have a neural correlate in the brain.

factualspeaker onlynovelty 3/4durability 2/4· Dwarkesh Patel

“the way we're going to learn how the brain works is just trying to first-principles engineer intelligence in AI. Then it just happens to be the case that each one of these things has a neural correlate.”