Zachary Lipton

wiki
, , , ,
1

Raj Reddy Associate Professor of Machine Learning at CMU, co-founder and CTO of Abridge, and one of the field’s most prominent critical voices — known equally for foundational work on distribution shift, model interpretability, and clinical ML, the Dive into Deep Learning textbook, and a sharp-edged blog that has shaped how the research community argues about its own practices.

Profile

Field Detail
Full name Zachary Chase Lipton
Date of birth August 19, 1985
Place of birth New Rochelle, New York, USA
Nationality American
Current institutions Carnegie Mellon University; Abridge
CMU role Raj Reddy Associate Professor of Machine Learning
Industry role Co-founder & CTO, Abridge
Research areas Distribution shift, model interpretability, clinical ML, NLP, ML methodology, robust/adaptive systems
PhD thesis Learning from Temporally-Structured Human Activities Data (UCSD, 2017)
PhD advisors Charles Elkan; Julian McAuley
Research group ACMI Lab (Approximately Correct Machine Intelligence)
Personal website zacharylipton.com
Blog approximatelycorrect.com
X / Twitter @zacharylipton
GitHub zackchase
Google Scholar MN9Kfg8AAAAJ

Overview

Zachary Chase Lipton is the Raj Reddy Associate Professor of Machine Learning at Carnegie Mellon University, where he directs the ACMI Lab, and co-founder and CTO of Abridge — a Pittsburgh-based healthcare AI company building ambient AI for clinical documentation that has raised over $300 million at a $5.3 billion valuation. His research spans ML methodology and theory, distribution shift, model interpretability, NLP, and clinical applications, and he is a co-author of Dive into Deep Learning (Cambridge University Press, 2023), an interactive open-source textbook adopted at over 500 universities in 70 countries. He is widely recognized for critical essays that have shaped community discourse — most notably “The Mythos of Model Interpretability” and “Troubling Trends in Machine Learning Scholarship” — and for the Approximately Correct blog, which combines technical depth with social criticism and occasional satire. He is also a jazz saxophonist who recorded a debut album in 2007 and studied under Branford Marsalis, and is the grandson of Issachar Miron, the Israeli-American composer who co-wrote “Tzena, Tzena, Tzena.”

Early Life & Education

Lipton was born on August 19, 1985, in New Rochelle, New York. He is the grandson of Issachar Miron (1920–2015), an Israeli-American composer best known for co-writing the Hebrew song “Tzena, Tzena, Tzena” (1941), which became an international hit following English adaptations by Pete Seeger and The Weavers in the 1950s. Lipton began playing the saxophone at age eight and developed into a serious jazz musician through his teens, receiving instruction from Steve Wilson, Bill McHenry, and Branford Marsalis, who praised his “rare dedication to often overlooked elements of musicianship” and “innate sense of curiosity and meticulous attention to detail.” As a high school senior, he was selected to perform in the inaugural New York Youth Symphony Jazz Orchestra alongside Slide Hampton and Jimmy Heath.

He enrolled at Columbia University, where he earned a B.A. in Mathematics and Economics (c. 2003–2007). During his senior year, he recorded a collection of original compositions on a debut jazz album, First Steps (2007), praised in All About Jazz as “an engaging introduction, musically genuine and well-conceived.” After graduating, he remained in New York for several years pursuing music professionally while doing web development on the side. He gradually transitioned toward computer science, eventually enrolling in the PhD program at UCSD.

During his PhD, Lipton interned at Amazon Core Machine Learning in Seattle (summer 2015, working on recommendation systems for Prime Video) and Microsoft Research (the following year), as well as a period living in India (summer 2014). He received his Ph.D. in Computer Science from the University of California, San Diego (AI Group) in 2017, advised by Charles Elkan and Julian McAuley. His dissertation, Learning from Temporally-Structured Human Activities Data, addressed sequence modeling and learning from time-series data produced by human activities.

Career

Amazon AI — Visiting Scientist / Research Scientist (2018–c. 2023)

Following his PhD, Lipton held a concurrent visiting/research scientist position at Amazon AI while taking up his faculty position at CMU. He had previously interned with Amazon Core Machine Learning during his doctoral work and continued this industrial affiliation after joining academia, contributing to research in NLP and ML systems.

CMU — Raj Reddy Associate Professor of Machine Learning (2018–present)

Lipton joined Carnegie Mellon’s Machine Learning Department in 2018, holding his primary appointment there alongside courtesy appointments in the Heinz School of Public Policy and Societal Computing. He directs the ACMI Lab (Approximately Correct Machine Intelligence), whose research covers robust ML under distribution shift, causality and representation learning, clinical ML, and the social impacts of AI systems. He is also appointed in the CMU Software and Societal Systems Department (S3D) as an affiliated faculty member. His named chair — the Raj Reddy Professorship — is named after the CMU AI pioneer and Turing Award laureate. He has previously held affiliations with Microsoft Research Redmond and Microsoft Research Bangalore.

Abridge — Co-founder & CTO (2023–present)

Abridge was founded in Pittsburgh in 2018 by Shivdev Rao (CEO), Florian Metze, and Sandeep Konam as an AI-powered ambient listening platform for clinical documentation. Lipton joined the company in early 2023, initially as Chief Scientific Officer, and subsequently became Co-founder and CTO. Under his technical leadership, Abridge’s platform — which listens to doctor-patient conversations and automatically generates structured clinical notes integrated with EHR systems such as Epic — has scaled to major health system partnerships. The company raised a Series C of $150 million (led by Lightspeed and Redpoint Ventures, early 2024) and a subsequent round of $300 million at a $5.3 billion valuation (2025), bringing total funding to well above $500 million. CEO Shiv Rao has said of Lipton: “Since meeting Zack five years ago, I have always been deeply impressed by his AI research, which pioneered the applications of many modern machine learning techniques to healthcare data.” Lipton frames the core technical challenge as not merely building accurate models but deploying them safely in live clinical environments — monitoring, evaluating, and adapting models against hallucination and distributional drift in high-stakes settings.

Key Contributions

  • “The Mythos of Model Interpretability” (Lipton; ICML 2016 Workshop; ACM Queue and Communications of the ACM, 2018) — A foundational critique arguing that “interpretability” in ML is underspecified, that its diverse motivations are sometimes mutually incompatible, and that common assertions (e.g., linear models are interpretable, deep networks are not) are often unjustified. Over 1,400 citations; widely cited in fairness, accountability, and transparency research and in ML ethics curricula.

  • “Troubling Trends in Machine Learning Scholarship” (Lipton and Steinhardt; ICML 2018 Debates; Communications of the ACM, 2019) — Catalogued four systematic failures in ML scholarship: conflating explanation with speculation, obscuring sources of empirical gains, mathematical obfuscation (“mathiness”), and linguistic misuse. Presented at ICML’s Machine Learning Debates session; became a frequently assigned reading in ML methodology courses and a touchstone for discussions of research rigor.

  • “Detecting and Correcting for Label Shift with Black Box Predictors” (Lipton, Wang, Smola; ICML 2018) — Introduced a practical method for detecting and correcting label shift (a common form of distribution shift in deployment) using only a pre-trained black-box classifier, without access to the model’s internals. Contributed to formalizing the label-shift problem and motivating a line of work on distribution shift correction.

  • “A Critical Review of Recurrent Neural Networks for Sequence Learning” (Lipton, Berkowitz, Elkan; 2015) — An early comprehensive survey of RNNs that clarified terminology and organized the then-nascent literature on sequence modeling; widely read as one of the first rigorous reviews of the area at the advent of the deep learning era.

  • Dive into Deep Learning (Zhang, Lipton, Li, Smola; Cambridge University Press, 2023) — A fully interactive open-source textbook integrating exposition, mathematics, and runnable Jupyter notebook code, supporting PyTorch, TensorFlow, and MXNet. Adopted at over 500 universities in 70 countries, including Stanford, MIT, Harvard, and Cambridge. Praised by Jensen Huang (NVIDIA CEO) and endorsed by leading researchers across the field. Lipton is the second-named co-author alongside Aston Zhang, Mu Li, and Alexander Smola.

  • ACMI Lab research on distribution shift and clinical ML — Sustained body of work on how ML models behave under natural and adversarial distribution shifts, their application to clinical time-series data and healthcare NLP, and the conditions under which offline-trained models can be trusted in deployment. Representative work includes papers on concept drift, label shift, and ML for ICU prediction tasks.

  • Approximately Correct blog (launched 2016) — A long-running blog bridging technical ML perspectives and social criticism, notable for essays on interpretability, NLP hype, conference culture, and AI ethics that reached wide audiences well beyond typical academic readership. Posts include “OpenAI Trains Language Model, Mass Hysteria Ensues” (2019) and “The Greatest Trade Show North of Vegas,” as well as the Superheroes of Deep Learning comic series.

Awards & Recognition

  • Raj Reddy Professorship — Named chair at CMU, honoring the Turing Award–winning founder of CMU’s AI program
  • Featured expert, New York Times — “What’s the Future for A.I.?” (quoted on the economic impact of AI)
  • Featured expert, CNBC — Quoted on OpenAI governance following the 2023 board shakeup
  • Axis 2024 feature, Axios — Quoted in “AI’s flawed human yardstick” on benchmark inflation
  • Dive into Deep Learning adopted at 500+ universities across 70 countries

Key Relationships

  • Shivdev Rao — CEO and original founder of Abridge; non-invasive cardiologist at UPMC; the physician-researcher collaboration that defines Abridge’s clinical grounding. Rao has credited Lipton’s healthcare ML research as a primary reason for their partnership.

  • Alexander Smola — Co-author of Dive into Deep Learning; VP/Distinguished Scientist at Amazon Web Services and one of the most prominent kernel methods and ML systems researchers; Lipton’s Amazon affiliation and D2L collaboration trace to this relationship.

  • Aston Zhang — First-named co-author of Dive into Deep Learning (now at OpenAI); the project originated from Amazon internal course material developed by Zhang, Li, and Smola, with Lipton joining to expand and publish it.

  • Charles Elkan — PhD advisor at UCSD; known for work on learning from imbalanced data and probabilistic modeling; shaped Lipton’s early training in rigorous ML theory and evaluation.

  • Julian McAuley — PhD co-advisor at UCSD; prominent in recommender systems and NLP; the combination of Elkan and McAuley reflects Lipton’s dual orientation toward theory and applied NLP.

  • Jacob Steinhardt — Co-author of “Troubling Trends in Machine Learning Scholarship”; professor of statistics at UC Berkeley and prominent voice in ML safety; the paper represents a shared intellectual concern about the epistemic standards of the ML research community.

Personal Style

Lipton occupies an unusual triple position in the AI ecosystem — active researcher, company builder, and field critic — and is reluctant to mute any of the three roles for the sake of the others. His X presence (@zacharylipton, ~66K followers) is the most distinctly humorous and provocative of any senior ML academic: posts range from precise technical arguments about benchmark integrity and distributional shift to satirical takedowns of hype cycles, startup valuations, and conference culture. His Digg vibe profile ranks him as the most humorous ML researcher in its top 35, with equal presence in the “provocative” category. The Approximately Correct blog, launched in 2016, established early that his criticism is not performed contrarianism but a consistent analytical stance about what the field gets wrong in its self-presentation and methodology. He frames the saxophone as a sustained identity rather than a past hobby — the “Relapsing :saxophone:” tag on his X bio reflects his own description of his relationship with jazz as something that periodically reasserts itself against his professional obligations.

References