Micah Goldblum

Zebra-CoT: A Dataset for Interleaved Vision-Language Reasoning

Teaching Pretrained Language Models to Think Deeper with Retrofitted Recurrence

Identifying and Evaluating Inactive Heads in Pretrained LLMs

Privacy-Preserving Mechanisms Enable Cheap Verifiable Inference of LLMs

Closing the Train-Test Gap in World Models for Gradient-Based Planning

Incoherent Beliefs & Inconsistent Actions In Language Models

LiveBench: A Challenging, Contamination-Limited LLM Benchmark

Small Batch Size Training for Language Models: When Vanilla SGD Works, and Why Gradient Accumulation is Wasteful

Far from the Shallow: Brain-Predictive Reasoning Embedding through Residual Disentanglement

Refusal Tokens: A Simple Way to Calibrate Refusals in Large Language Models

Adaptive Retention & Correction: Test-Time Training for Continual Learning

Gemstones: A Model Suite for Multi-Faceted Scaling Laws

Style Outweighs Substance: Failure Modes of LLM Judges in Alignment Benchmarking

Just How Flexible are Neural Networks in Practice?

vTune: Verifiable Fine-Tuning for LLMs Through Backdooring

Refusal Tokens: A Simple Way to Calibrate Refusals in Large Language Models

Hidden No More: Attacking and Defending Private Third-Party LLM Inference

Data Brittleness Estimation with Self-Supervised Features

A Simple Baseline for Predicting Future Events with Auto-Regressive Tabular Transformers

LLMs Boost the Performance of Decision Trees on Tabular Data across Sample Sizes