Oldest Papers
Ordinary people
rayne fisher-quann
Substack·2026
A response to a bad article, a bad book, and a bad publication.
wrote a long one about one of the worst essays I've ever read. also about the Free Press, decorum politics, nellie bowles, Minneapolis, and the paradox of trying to criticize a political force whose ultimate goal is the proliferation of discourse itself. fuck ICE.
No ratings yet
View paper →Pandora’s Box
Unknown Author
BJPsych International·2026
No abstract available
No ratings yet
View paper →Centrist imaginations
rayne fisher-quann
Substack·2026
A response to a bad article, a bad book, and a bad publication
★ 5.0 (1)
View paper →The dynamic between math engagement and perceived teacher expectations during adolescence: Insights on educational inequalities in math achievement
Wen Wen, Christina L Scanlon, Ming-Te Wang
Child Development·2026
Abstract
This three-wave longitudinal study (2016 to 2018; N = 1,749; Mage = 13.83, SD = 1.67; 50.1% male; 67.8% White, 32.2% Black) in the US examined the reciprocal association between math engagement and perceived teacher expectations over time and their impact on math achievement and explored racial and socioeconomic status differences in these dynamics. Results showed that both math engagement and perceived teacher expectations positively influenced math achievement. Math engagement predicted higher next-year perceived teacher expectations (b = 0.204, SE = 0.050, p < .001), but the reverse was not observed. This effect was stronger among White (vs. Black) and high-SES (vs. low-SES) students. These findings underscore student agency in shaping teacher–student interactions and persistent racial and SES disparities in educational experiences.
No ratings yet
View paper →The Shape of Beliefs: Geometry, Dynamics, and Interventions along Representation Manifolds of Language Models' Posteriors
Raphaël Sarfati, Eric Bigelow, Daniel Wurgaft, Jack Merullo, Atticus Geiger, Owen Lewis, Tom McGrath, Ekdeep Singh Lubana
arXiv·2026
Large language models (LLMs) represent prompt-conditioned beliefs (posteriors over answers and claims), but we lack a mechanistic account of how these beliefs are encoded in representation space, how they update with new evidence, and how interventions reshape them. We study a controlled setting in which Llama-3.2 generates samples from a normal distribution by implicitly inferring its parameters (mean and standard deviation) given only samples from the distribution in context. We find representations of curved "belief manifolds" for these parameters form with sufficient in-context learning and study how the model adapts when the distribution suddenly changes. While standard linear steering often pushes the model off-manifold and induces coupled, out-of-distribution shifts, geometry and field-aware steering better preserves the intended belief family. Our work demonstrates an example of linear field probing (LFP) as a simple approach to tile the data manifold and make interventions that respect the underlying geometry. We conclude that rich structure emerges naturally in LLMs and that purely linear concept representations are often an inadequate abstraction.
No ratings yet
View paper →