Tomayto Tomahto

Keyon Vafa (postdoc at Harvard) studies the implicit world models that generative models learn (or at least are constrained into representing). How do we make sense of these models and how do they make sense of us? Experimentally, answering this question often looks like giving the model anything and everything: cookie ingredients, language, shrimp taxonomies, protein structures, a sample of ABBA’s discography—and models will try to make sense of it. Under a methodologically unified view of an incredibly variegated world with diverse objects, Keyon has tried to figure out if models can learn the map of Manhattan, the laws of physics, the political ideology of Senators, and more. This is partially an interdisciplinary agenda and partially an anti-disciplinary agenda.
In this episode, we cover Keyon's research on world models, human and AI interaction, machine learning methods to study the gender wage gap, the implications of thinking with and through predictability vs. causality, and documentary film making. It's a blast, as always, to be thinking methodologically and free ourselves from the bounds of disciplinary objects!
⁠Keyon Vafa's Website⁠
⁠Senate Speech presentation⁠
⁠Taxi rides⁠
⁠CAREER paper⁠
Indexing Political Persuasion: Variation in IRAQ vowels
AlphaFold
David Donoho on frictionless reproducibility
Frederick Wiseman
Measuring the predictability of life outcomes with a scientific mass collaboration

Across Amanda Anderson’s scholarship, she notes a mutually beneficial and complex relationship between the aesthetic and the political. There’s a “political potency” to art that arises, one might guess, from the fact that “the humanities use the aesthetic to express political commitment, rather than to run from it.” Far from weakening our understanding of the political, an aesthetic engagement can be an "intensification" of political insight. Nevertheless, not every aesthetic project must be political or vice versa. Not every political insight would benefit from aesthetic engagement. The relationship between the aesthetic and the political needn’t be a co-constitutive one, but rather a mutually beneficial bond. That is, we can leverage our aesthetic engagement with the political to more precisely illuminate the relationship between the systemic and the lived, between the collective and the individual, between how power is felt and how it’s institutionalized. This episode will move through several of Amanda’s scholarly projects—Humanities Theory, Bleak Liberalism, and debates over the importance of an education in literary studies. What should we teach in English classrooms and why? Are humanities classrooms bastions of pluralism and liberal ideals—or do they merely purport to value ideological heterogeneity? Must we commit (and why) to liberalism rather than progressivism or neoliberalism? Amanda addresses these questions and more with striking precision and a plethora of sources.
Amanda Anderson is the Andrew W. Mellon Professor of Humanities and English at Brown University. From 2015 - 2026 she is the Director of the Cogut Institute for the Humanities.

⁠Amanda Anderson research profile ⁠
⁠Meeting Street ⁠
⁠Humanities Theory ⁠
Cruel Optimism 
Slow Violence 
Saidiya Hartman (critical fabulation) 
Bleak Liberalism
Psyche and Ethos 
⁠Music by Blue Dot Sessions ⁠(https://www.sessions.blue/)

Ultimately, this episode is about science and scholarship. As Sean says,  “understanding something as well as you can in science means that you need to confront the data and be pushed out of your comfort zone.” I find it counterintuitive but true: this episode shows us that theoretical physics and indeed science pushes us into the subjunctive. It’s our job as scholars to think beyond  what’s given, beyond what’s happening right now around us, and think about what could happen, perhaps what would happen if certain constraints were lifted.
If we suffered a mass extinction, what would life look like? If the mouth were configured differently, how would phonetic change have been different from the beginning? What about the uniformitarian hypothesis? If a language dies out and a new hybrid language forms, what are the possibilities and impossibilities? And then what happens when we think about this space of possibilities combinatorially vs. probabilistically vs. normatively?
Among other things, Sean and I discuss the romance of the university, the merits of interdisciplinarity, his blog posts from 20+ years ago on Zizek, language, and metaphor—we inevitably touch on AI and writing—and, of course, we discuss what it means to host podcasts and present public scholarship.
Sean Carroll, the host of Sean Carroll's Mindscape, is the Homewood Professor of Natural Philosophy at Johns Hopkins University. He is the author of several books including The Biggest Ideas in the Universe: Space, Time, and Motion, and Something Deeply Hidden: Quantum Worlds and the Emergence of Spacetime.
https://www.preposterousuniverse.com/ 
The universe is structured like a language
From experience to metaphor by way of imagination
Holiday message 2025
The Universe as a Quantum Computer
The Library of Babel 
David Krakauer
Maine senate primary
Max Weber 
⁠Music by Blue Dot Sessions ⁠(https://www.sessions.blue/)

We've been told time and time again that we need to understand data in context: it's an ethical imperative. Not every language gets an LLM; not every population fully understands a technology that's deployed in their community with or without everyone's consent; and certainly we're told that we will make better, safer conclusions with our data if we understand the context. John DeNero looks at things differently: instead of an ethical imperative for understanding data in context, John talks about a structural one. For example, accurately translating language necessitates understanding the context. It's almost as if he read a bunch of French critical theory, thought about deconstruction, and realized that a structural imperative has an ethical valence as well—and vice versa. It's not a paradox, it's deconstruction.
This interview covers John's work as a professor of data science and computer science, his experience as a senior research scientist at Google Translate, thoughts on AI and language, and keeping up with the slang of today's youth.
John DeNero is the Faculty Director of Data Science Undergraduate Studies (DSUS) and Associate Teaching Professor in the UC Berkeley EECS department. He is the co-founder and Chief Scientist at lilt.
John's website
Google Scholar
A Class-Based Agreement Model for Generating Accurately Inflected Translations
⁠Music by Blue Dot Sessions ⁠(https://www.sessions.blue/)

This episode is dedicated to MukhammadAziz Umurzokov and Ella Cook, the two Brown University students who passed away on December 13th, 2025.

In language-centric fields we privilege the speaker. Linguistics looks at spoken or signed utterances; linguistic anthropology does as well. But Michael Berman looks at listening, which for him is a process wherein you limit or shift your language practices so as to avoid being generated as a certain type of person (often within a hierarchical relationship). That’s listening. It's about avoiding (or not) taxonomy, stereotypes, perception, and it necessitates an understanding of the power that our ears have. This episode cannot be reduced to a few thematic elements: Michael and I discuss listening, semiotics, C.S. Peirce, suffering and compassion, critiques of linguistics and other sciences, the implicit economic models undergirding scholarship, and his fieldwork in Japan—among other things. I’m struck by how much ground we cover, and yet we make a limited number of rhetorical and analytic moves. Whether we’re talking about what constitutes listening, language ideology, religion, etc.—we’re always taking the minuscule and making it representative (or symptomatic) of something bigger. Maybe that’s a paranoid reading, but I think it’s useful in the context of our conversation. What appears as an individual assessment of language is in fact a societally-engineered and collectively-upheld assessment. What appears as a certain niche orientation to data turns out to be symptomatic of widespread abuses of scientific frameworks. And, as Michael will remind us, the creation of categories and production of knowledge has effects. So let’s pay attention. 
This episode took inspiration from the questions that Jonathan Rosa asked in his episode on Tomayto Tomahto a year ago. Before listening to Michael, I encourage listening to Jonathan’s episode if you haven’t already. 
Michael Berman
C.S. Peirce
Jonathan Rosa’s episode 
Toward a linguistic anthropological approach to listening: An ear with power and the policing of “active listening” volunteers in Japan
Religion overcoming religions: Suffering, secularism, and the training of interfaith chaplains in Japan
Forms of the Affects 
“Why The Problem Isn’t Single-Parent Families” 
Why Is There So Much Fraud in Academia?
⁠Music by Blue Dot Sessions ⁠(https://www.sessions.blue/)
This episode was written, edited, and produced by Talia Sherman. All artwork by Maja Mishevska.