Feature learning, alignment and the linear representation hypothesis for steering & monitoring LLMs @FieldsInstitute
Feature learning, alignment and the linear representation hypothesis for steering & monitoring LLMs  @FieldsInstitute
Uploaded September 2026 | Updated November 2026, 2 days ago
Speaker: Mikhail Belkin, University of California, San Diego
Date: September 2, 2026
Abstract: fields.utoronto.ca/talks/Feature-learning-alignment-and-linear-representation-hypothesis-steering-and-monitoring-LLMs

2026-2027 Mathematical AI Seminar: fields.utoronto.ca/activities/26-27/mathai

Centre for Mathematical AI: fields.utoronto.ca/centres/centre-mathematical-ai
Feature learning, alignment and the linear representation hypothesis for steering & monitoring LLMsDetailed assessment of calculating drag force with quantum computersHomoclinic orbits through beyond all order phenomenaScaling limits of random planar mapsLorentzian Optimal Transport and Almost-SplittingFrom Goal-Based Portfolios to Insurance ALM: Implementing SDDP with Open-Source ToolsWeak interactions for estimating the complete CHSH parameter from each entangled photon pairRigidity of {beta}-Mather function for generalized standard mapsAn Analytical Framework for Reverse Stress TestingQuasi-birth-and-death processes evolving within phylogenetic trees: Modelling traits of speciesRG for percolatorsUnique ergodicity of branched covers of translation surfaces
Fields Institute |

Feature learning, alignment and the linear representation hypothesis for steering & monitoring LLMs

SHARE TO X SHARE TO REDDIT SHARE TO FACEBOOK WALLPAPER