Mark Zobeck

I'm a doctor who does research. I've annoyed DOZENS of statisticians. Here are 5 proven ways to exasperate them. #medtwitter #epitwitter #statstwitter

Researchers from MIT, Harvard, UCSF, and Boston University CONFIRM that EPIC SUCKS and is a MONOPOLY, not because of superiority but because of our stupid legislators, and is causing: Stagnation of Innovation Inequities in Data Access Compromised Patient Care Increased cost of Care Epic Systems has established an extraordinarily dominant position in U.S. healthcare IT, controlling over half of acute care hospital beds and dominating academic medical centers. It functions as the de facto gatekeeper for health data belonging to hundreds of millions of Americans. Its market power stems not from superior technology but from structural lock-in mechanisms that make switching (defection) extremely expensive and difficult for hospitals and health systems. The government has enabled the concentration of the “digital backbone” of American healthcare in a single private vendor, and we're paying for it dearly. Link to the study in the next post.

Extremely good.

New high-effort article "Why Creativity Cannot Be Interpolated" co-written with Dr. Jeremy Michael Budd. Yes the name is a pun on the famous book by @kenneth0stanley! The counterintuitive thesis (corollary of Kenneth's research): - Intelligence and agency are orthogonal to creativity - and sometimes actively hostile to it. - Genuine creativity is impossible without deep understanding and creativity without understanding is "slop". The strangest property of LLMs: within a single frame they seem to comprehend so deeply, yet they possess no perspective of their own. Like the blind men and elephant parable, each report is accurate, yet none integrates. We call this "frame-dependent" understanding, and it will change how you think about AI creativity. We started writing this 2 years ago, and this is our distilled understanding of AI creativity in 2026.

Oof. Always draw the DAG.

@PavlosMsaouel @NatureComms @MDAndersonNews @PGrivasMDPhD @montypal @VivekSubbiah @timheff5 @JadChahoud @KidneyCancer @kidneycan @neerajaiims @DrJeffreyGraham @IamLinghua Fascinating work. Impressive as always! Congrats!

1/5 Our new @NatureComms study reports the first-ever dedicated clinical trial for #RenalMedullaryCarcinoma (RMC): phase II trial of #nivolumab + #ipilimumab and what it taught us about hyperprogression to immunotherapy: nature.com/articles/s4146… #endRMC #MsaouelLab @MDAndersonNews

@PavlosMsaouel @kaydaustin @ERPlimackMD @crisbergerot @NazliDizman @BradMcG04 @g_procopio_ @maughanonc @salvolarosa @LabGenovese @JAMouabbi @IamLinghua @OAlhalabiMD @perelli_luigi @rahulshethmd @ChadTangMD @TiansterZhang @QingZhangLab @BCottaMD @Daniel_J_George @DrDanielHeng @priyaraomd @UroDocAsh @drmehrarohit @ED_PhD_ @DanielFrigo @rovingatuscap @katy_beckermann @shilpaonc @AmandaNizamMD @ZiadBakouny @eric_facs @CCoarfa @lalaniMD @WallisCJD @HindRafei @SGoswamiMDPhD @mjmoussa_ @AlbigesL @oncopathologist This is fascinating. Excellent work as always, Pavlos. Congrats!

@kaydaustin @ERPlimackMD @crisbergerot @NazliDizman @BradMcG04 @g_procopio_ @maughanonc @salvolarosa @LabGenovese @JAMouabbi @IamLinghua @OAlhalabiMD @perelli_luigi @rahulshethmd @ChadTangMD @TiansterZhang @QingZhangLab @BCottaMD @Daniel_J_George @DrDanielHeng @priyaraomd @UroDocAsh @drmehrarohit @ED_PhD_ @DanielFrigo @rovingatuscap @katy_beckermann @shilpaonc @AmandaNizamMD @ZiadBakouny 5/5 Clinical evidence: Sacituzumab govitecan in 4 heavily pretreated RMC pts → 1 PR (~5.3 mo) + 2 SD; median PFS 2.9 mo. Early, small-n but proof that biology-guided targeting can bend the curve for RMC.  Next: more RMC-specific trials based on these biological insights.

1/5 Our new @CellRepMed study delivers the largest #RenalMedullaryCarcinoma molecular profiling via @BostonGeneCorp + translational testing designed to uncover targetable vulnerabilities: cell.com/cell-reports-m… #MsaouelLab #endRMC #rarecancers

🗞️ Breaking: @AfricaCDC and the Republic of Ghana, in collaboration with @TexasChildrens, have launched “A New Day for Children with Sickle Cell Disease” at #UNGA80. This initiative will integrate SCD into primary health care, strengthen early detection, and expand access to affordable medicines—helping more children survive and thrive. Read more: ow.ly/cqwQ50X1kpJ #AfricaCDC #NewPublicHealthOrder

"AI isn't replacing radiologists" good article Expectation: rapid progress in image recognition AI will delete radiology jobs (e.g. as famously predicted by Geoff Hinton now almost a decade ago). Reality: radiology is doing great and is growing. There are a lot of imo naive predictions out there on the imminent impact of AI on the job market. E.g. a ~year ago, I was asked by someone who should know better if I think there will be any software engineers still today. (Spoiler: I think we're going to make it). This is happening too broadly. The post goes into detail on why it's not that simple, using the example of radiology: - the benchmarks are nowhere near broad enough to reflect actual, real scenarios. - the job is a lot more multifaceted than just image recognition. - deployment realities: regulatory, insurance and liability, diffusion and institutional inertia. - Jevons paradox: if radiologists are sped up via AI as a tool, a lot more demand shows up. I will say that radiology was imo not among the best examples to pick on in 2016 - it's too multi-faceted, too high risk, too regulated. When looking for jobs that will change a lot due to AI on shorter time scales, I'd look in other places - jobs that look like repetition of one rote task, each task being relatively independent, closed (not requiring too much context), short (in time), forgiving (the cost of mistake is low), and of course automatable giving current (and digital) capability. Even then, I'd expect to see AI adopted as a tool at first, where jobs change and refactor (e.g. more monitoring or supervising than manual doing, etc). Maybe coming up, we'll find better and broader set of examples of how this is all playing out across the industry. About 6 months ago, I was also asked to vote if we will have less or more software engineers in 5 years. Exercise left for the reader. Full post (the whole The Works in Progress Newsletter is quite good): worksinprogress.news/p/why-ai-isnt-…

This is helpful, but we're still far away from causal reasoning, in the sense of do(x)->y. Grounding reasoning in the real world is a must. RLHF will only ever be a cartoon sketch of causation.

It is astounding to me that this is not common knowledge among the machine learning community.

@kareem_carr Congrats!

submitted the final draft of my dissertation last night. feeling extremely proud but also extremely exhausted.

In our latest work, we demonstrate that risk estimates for patients are HUGELY uncertain due to model uncertainty, data uncertainty, and population uncertainty. Even when based on large sample sizes. @laure_wynants @ESteyerberg arxiv.org/abs/2506.17141

🫣#rstats

@ErwanLamy1 @eliasbareinboim I'll add that the two are complementary: The FEP depends on an agent's generative model of the data. An agent should use causal inference methods to build a generative model that causally reflects the generative process, which it can use to further minimize free energy.

@eliasbareinboim Thanks for the very clear answer.

In a recent work (causalai.net/r136.pdf), we examined whether LLMs are potential sources of probabilistic knowledge (rung 1 of Pearl's hierarchy), which led to the benchmark at llm-observatory.org. The answer was no, which was surprising and poses fundamental challenges for various downstream tasks and key capabilities (including explanation, decision-making, generalization, safety, and learning) given that inferences about interventions (rung 2) and counterfactuals (rung 3) build on rung-1 knowledge. Many have asked me about the potential of LLMs over the past year or so. In short: they could become extraordinary repositories of knowledge, à la Internet or Wikipedia. But when it comes to the broader ambitions of AI, there’s still a good way to go; the interplay between language and causality remains largely unmapped and only rudimentarily understood. I hope more young researchers will join the effort to tackle these generational challenges. Despite all the hype, we still need foundational principles for a better science of intelligence, one that integrates language, causality, and other essential components.

NCI payline: 4% Likelihood of alien invasion in November: 7% The likelihood of getting a NCI R01 grant is lower than that of an alien invasion during the month of November. Let that sink in. @SyedAAhmad5 #CancerResearch

I cant get over the fact that so many engineers still dont grok the fundamentals of what an LLM is. Repeat after me: its just pattern matching, it doesnt "know" anything

@PavlosMsaouel @AlexSherryMD @ErikVanZwet @ebludmir @d_shapiro1 @PGrivasMDPhD @montypal @OncHahn @JadChahoud @f2harrell @neerajaiims @DrJeffreyGraham @Richard_D_Riley @chubbyemu This is fascinating, Pavlos. Wonderful work as always!

1/5 How reproducible are phase III oncology trials? 📊 With @AlexSherryMD @ErikVanZwet @ebludmir et al. We analyzed 632 RCTs (~496,000 pts). The data offer both reassurance and caution for clinicians and researchers. pubmed.ncbi.nlm.nih.gov/40628175/ 👇#MsaouelLab

Very interesting paper on causal inference from statistical models; perhaps more actionable than most causal inference papers: tandfonline.com/doi/full/10.10… #Statistics

Timeless advice

@PavlosMsaouel @Soum_Roy_RadOnc @_MiguelHernan @MDAndersonNews @OAlhalabiMD @OncHahn @f2harrell @JadChahoud @PGrivasMDPhD @maxinesun @DrJeffreyGraham @ebludmir @UroDocAsh Agreed! I'll add that clinicians use informal causal inference reasoning on a daily basis (e.g. what is causing a patient's fever?). Sometimes we do it well. Many times we do it poorly. Learning about formal causal inference can help improve clinical reasoning.

@Soum_Roy_RadOnc @_MiguelHernan @MDAndersonNews @OAlhalabiMD @OncHahn @f2harrell @JadChahoud @MarkZobeck @PGrivasMDPhD @maxinesun @DrJeffreyGraham @ebludmir @UroDocAsh Invaluable on a daily basis & used regularly by our faculty & statisticians. All clinicians should have at least some familiarity with causal inference & most data scientists should be ready to use its tools as needed. Concerns of misuse are similar to concerns for Bayes or freq.

We are excited to host @_MiguelHernan in Houston @MDAndersonNews next week to give this year’s Melvin L. Samuels lecture (hybrid; zoom info in photo). Looking forward to the synergies that will emerge from his visit to advance rigorous causal inference methods across oncology.