Evolution didn't select against aging mainly because the ancestral hazard rate was too high for it to matter.
Kimmel's first-principles argument: even without aging, ancestral humans faced such high day-to-day mortality (predation, infection, accidents) that very few individuals ever lived long enough for longevity-extending alleles to accumulate meaningful selection pressure. The 'gradient signal' for longevity flowing back to the genome was weak simply because so few organisms reached old age.
evolutionary-theory-of-aging
Kin selection can actively penalize longevity, not just fail to reward it.
If an individual lives longer without having aging fully solved, their marginal calorie contribution to the group falls short of what two new twenty-year-olds could contribute in their place. A population that carries many partially-aged, lower-fitness individuals can be a net drag on genome propagation, giving evolution a positive reason to favor turnover over longevity.
evolutionary-theory-of-aging
Evolution's own optimizer had a limited 'budget' (mutation rate x population size), and that budget was spent overwhelmingly on infectious-disease resistance rather than longevity.
Kimmel frames mutation rate as evolution's step size and population size as its batch size for parallel search. Infectious disease was the dominant force shaping early human population demographics, so most of the available evolutionary 'gradient' was directed there rather than toward extending lifespan, even setting aside whether longevity was selected for at all.
evolutionary-theory-of-aging
Peak fluid intelligence and peak scientific achievement cluster around age 25-30 likely because that was the age of the bulk of the reproducing population under selection, not because of an intrinsic cognitive ceiling.
Kimmel's 'pet hypothesis': if very few people in ancestral populations lived past 30-40, there was little selection pressure to preserve fluid intelligence beyond that age. This offers a biological explanation for the historical pattern (Newton, von Humboldt, and other 'annus mirabilis' careers) of major discoveries clustering in scientists' twenties and thirties across very different cultures and eras.
evolutionary-theory-of-aging
NewLimit's bet is that aging is largely explained by degradation of the epigenome, the transcription-factor 'marks' that tell cells with identical genomes to behave differently.
Transcription factors act like orchestra conductors, binding DNA and turning genes on or off without performing functions directly themselves. As the epigenome shifts with age, cells lose access to the right genetic programs at the right times, making them less resilient to disease. NewLimit's approach is to find transcription-factor combinations that push the epigenome back toward its younger state without changing what type of cell it is.
epigenetic-reprogramming
Yamanaka's reprogramming discovery was tractable with brute-force screening only because success was cheaply visible and self-amplifying; aging reprogramming has neither property, which is why NewLimit needs predictive models instead.
Yamanaka could screen 24 transcription factors down to four because his readout (a fibroblast becoming a stem cell) was visually obvious and low-efficiency successes (as rare as 1 in 100,000 cells) still grew into visible colonies. Old and young cells of the same type look almost identical by comparison, and successful 'young' cells don't proliferate to reveal themselves, so aging reprogramming requires single-cell sequencing plus models trained to predict effects rather than visual screening.
ai-for-drug-discovery
The combinatorial space of candidate reprogramming factors (roughly 10^16 combinations from ~2,000 transcription factors) is far larger than all single-cell sequencing ever performed, which is why a predictive model of perturbation effects is the only tractable path to finding good combinations.
NewLimit trains models on sparse experimental sampling of transcription-factor combinations and their effects on cell state, then uses the model to predict in silico which untested combinations are most likely to revert an aged cell to a young state, treating the search as a generative sampling problem analogous to base-model pretraining followed by an RL-style value objective in LLMs.
ai-for-drug-discovery
The two dominant nucleic-acid delivery methods, lipid nanoparticles and viral vectors, both face hard physical or immunogenicity ceilings, and Kimmel expects the eventual solution to be engineered living cells that patrol the body and release payloads only when triggered.
Lipid nanoparticles must physically survive transit from the bloodstream to a specific target cell without fusing elsewhere; viral vectors are inherently at least somewhat immunogenic and don't naturally infect every cell type. Kimmel's long-run ("2100") bet is that delivery will instead be solved the way the immune system already solves it: engineered T- or B-cell-like cells with sensing circuitry (an AND-gate logic) that release a genetic payload only in the right tissue at the right time, and that can persist in the body for years.
drug-delivery
Rejuvenating even a single cell type or organ produces disproportionate, body-wide health benefits, because organs function as interconnected endocrine signaling systems.
Kimmel cites liver-transplant data (patients receiving younger livers show reduced risk across multiple unrelated diseases and better overall survival, not just improved liver-specific metabolism), bone-marrow transplant case histories, and a single-gene (TFAM) knockout in mouse T cells that dramatically shortens lifespan, as evidence that partial, single-tissue reprogramming could still deliver broad systemic health benefits well before every cell type is addressable.
epigenetic-reprogramming
Eroom's Law (biopharma R&D cost per new drug rising steadily since the 1950s) persists because, unlike AI scaling, biotech investment doesn't compound: expertise in synthesizing a molecule for one gene doesn't transfer to knowing which gene to target for the next disease.
Kimmel argues the hard part of drug discovery isn't making a molecule that hits a target, it's figuring out what to target in the first place, and that knowledge doesn't generalize across diseases the way general-purpose model scaling generalizes across tasks. He frames NewLimit's virtual-cell modeling approach as an attempt to import the compounding-returns property of ML scaling into biotech.
biotech-economics
Current insurance reimbursement, built around a roughly 3-4 year average churn between insurers, is structurally mismatched to durable, potentially one-time therapies whose benefits accrue over a decade.
No individual insurer is economically incentivized to pay for a therapy whose main benefit shows up five years after dosing if the average patient switches plans well before then. Kimmel expects a shift toward pay-for-performance reimbursement (spreading cost over the years a therapy demonstrably keeps working) and toward direct-to-consumer models like Eli Lilly's LillyDirect, especially for medicines patients feel the benefit of directly rather than ones only a physician pushes.
biotech-economics
Roughly a third of all Medicare spending occurs in a patient's final year of life, suggesting therapies that prevent acute late-life decline could reduce total healthcare spend even as they add new drug costs.
Kimmel argues that because pharmaceuticals are the one part of the healthcare system that reliably gets cheaper over time (drugs eventually go generic while remaining effective), shifting more of the health burden from late-life crisis administration toward earlier, preventive reprogramming medicines should, on net, lower total healthcare spending as a share of GDP rather than raise it.
biotech-economics
Books referenced
Through the Looking-Glass - Lewis Carroll - source of the Red Queen image Kimmel uses for pathogen-host and host-immune arms races, where organisms must keep evolving new defenses just to stay in place
Media referenced
Super-SOX - paper - Sergiy Velychko's paper showing a mutated, synthetic SOX2 factor reprograms cells to pluripotency more efficiently than the natural Yamanaka factors, evidence that natural transcription factors are not optimal for medicine design
Companies
NewLimit - Kimmel's company; develops epigenetic reprogramming medicines using combinations of transcription factors to revert aged cells toward a younger state
Roche / Genentech - cited as a counterexample to the trend of large pharma outsourcing early discovery - Roche's 2013 Genentech acquisition kept deep in-house R&D under Aviv Regev
Eli Lilly - LillyDirect cited as an early example of pharma moving to direct-to-consumer distribution for incretin mimetics, a model Kimmel expects to spread to longevity medicines
Arc Institute - home of Luke Gilbert's work showing a single targeted epigenetic edit can keep cells dividing 400+ times over years, evidence that epigenetic reprogramming can be durable
Techniques and frameworks
Epigenetic reprogramming (Yamanaka factors) - Shinya Yamanaka's discovery that four transcription factors can revert an adult cell to an embryonic stem cell state; NewLimit's whole platform is built on finding narrower factor combinations that revert age without reverting cell identity
Perturb-seq - single-cell technique (2016, multiple labs) that pairs a genetic perturbation with an RNA-sequencing readout of the resulting cell state, the core wet-lab primitive NewLimit uses to generate training data
Kin selection - used to argue evolution can be actively selected against longevity, since a population's total 'genome ROI' can be higher from turnover to younger individuals than from keeping a partially-aged individual alive longer
Red Queen hypothesis - explains why humans never evolved their own antibiotics: any host or pathogen defense triggers rapid counter-evolution by the other side, so slow-mutating mammals can't keep pace with fast-mutating microbes
CAR-T therapy - cited as a working precedent for engineering the immune system's detection machinery while leaving its payload-delivery machinery untouched, a template Kimmel thinks nucleic-acid delivery will eventually follow
Evolution strategies - Kimmel's analogy for how natural selection resembles a gradient-free ML optimizer: random parameter (mutation) perturbations plus selection approximate a gradient step, explaining why small genetic edits can drive large phenotypic changes
Summary
Jacob Kimmel, president and co-founder of NewLimit, joins Dwarkesh Patel to make the evolutionary case for why aging is fixable and to walk through NewLimit's technical bet on epigenetic reprogramming. The first third of the conversation is a first-principles argument for why evolution never optimized for longevity: ancestral hazard rates were so high that few individuals lived long enough for longevity alleles to matter, kin selection can actively penalize keeping partially-aged individuals alive over replacing them with younger ones, and evolution's limited mutation-rate and population-size "budget" was spent disproportionately on infectious-disease resistance. Kimmel extends the same argument to intelligence, floating the hypothesis that peak fluid intelligence clusters around age 25-30 because that was roughly the age of the bulk of the ancestral reproducing population, not because of any hard cognitive ceiling.
The conversation moves into deep biology: why humans never evolved their own antibiotics (the Red Queen dynamics that favor pathogens' massive population sizes and high mutation tolerance over mammals' slow-evolving genomes), how gene duplication lets evolution repurpose existing genes for new threats without breaking their original function, and a detailed walkthrough of TRIM5alpha, a human gene that once protected against an HIV-like virus before losing that function while fighting off a now-extinct endogenous retrovirus - and which researchers can re-edit today to restore HIV resistance.
Kimmel then explains NewLimit's actual platform: using transcription factors, the "orchestra conductors" that tell cells with identical genomes to behave differently, to remodel the aged epigenome back toward a younger state without changing a cell's identity. He contrasts this with Yamanaka's original four-factor reprogramming breakthrough, arguing Yamanaka's brute-force screening worked only because success was cheaply visible (a color change, colony growth) and self-amplifying (rare successes still multiplied into visible colonies) - conditions aging reprogramming lacks entirely, since old and young cells of the same type look almost identical and don't self-select. That gap is why NewLimit trains predictive models on Perturb-seq-style data (transcription-factor combinations paired with resulting single-cell RNA states) to search a combinatorial space of roughly 10^16 possible factor combinations that could never be tested directly.
A long middle stretch covers delivery: lipid nanoparticles and viral vectors are the two dominant methods for getting nucleic-acid medicines into cells, and both face hard physical or immunogenicity ceilings. Kimmel's most speculative bet is that by 2100 delivery will instead be solved with engineered living cells modeled on the immune system - cells that patrol the body, sense specific signals through AND-gate logic, and release a genetic payload only where and when needed, the same two-component architecture CAR-T therapy already exploits for cancer. He backs the case for even partial, single-tissue reprogramming mattering with liver-transplant and bone-marrow transplant data showing outsized, body-wide health benefits from rejuvenating just one organ, since organs function as interconnected endocrine signaling systems.
The final third turns to biotech economics. Kimmel diagnoses Eroom's Law (the inverse of Moore's Law: rising cost per new drug since the 1950s) as arising because biotech expertise doesn't compound the way AI model scaling does - knowing how to synthesize a molecule for one target doesn't tell you what to target for the next disease, which is the actual hard part of drug discovery. He closes with a case that insurance reimbursement structures, built around a roughly 3-4 year average churn between payers, are poorly matched to durable therapies whose benefits accrue over a decade, predicting a shift toward pay-for-performance and direct-to-consumer models like Eli Lilly's LillyDirect, and arguing that preventive reprogramming medicines should ultimately shrink rather than grow total healthcare spending, since about a third of all Medicare costs are concentrated in a patient's final year of life.
Notable Quotes
"We should still be able to treat it in the best animal models of that disease... For the majority of pathologies, we just don't have many of those examples." - Jacob Kimmel
"You can think about it as a modular system that evolution's already gifted us." - Jacob Kimmel
"If you were to get the question of, 'When would you like to be born as a patient?' you always want to be born as close to today as possible." - Jacob Kimmel
"Whenever we're trying to cure infectious diseases, we just have to deal with, 'Fuck, viruses have been evolving for billions of years... it's so hard.' Then whenever we're trying to do something else, we're like, 'Fuck, the immune system has been evolving for billions of years... how do we get past it?'" - Dwarkesh Patel
"Most of the risk is not in how to make an antibody to treat my particular target, it's in figuring out what to target in the first place." - Jacob Kimmel