Entity · technique

Textual Gradient Optimization

techniqueactivetextual-gradient-optimization-c99ffc03·1 events·first seen May 26, 2026

Aliases: Textual Gradient Optimization

Co-occurring entities

MGDA Multi-Task Learning LLM-as-a-Judge Spearman's rho PCGrad

More like this (12)

TextGrad Integrated Gradients behavioral-gradient validator TextImage Augmentation Gradient-Guided Reward Optimization Policy Gradient Methods Mask-Aware Policy Gradients for Diffusion Language Models Bias Leaves a Gradient Trail Latent Trajectory Discrimination for AI-Generated Text Detection stochastic gradient ascent Evolved Policy Gradients Gradient Labs

Recent events (1)

5arXiv · cs.CL·May 26, 2026·source ↗

Failure Modes of Multi-Objective Prompt Optimization for LLM Judges

This paper investigates multi-objective prompt optimization for LLM-as-judge systems, testing five decomposition modes of textual gradient optimizers across varying levels of cross-task information sharing. In 6 of 10 configurations, optimization fails to improve over the initial prompt, with gradient specificity dropping 59% when multiple criteria are processed jointly. The authors identify two separable failure modes: gradient dilution at optimization time and instruction interference at inference time. These findings constrain the design space for customizing LLM judges via textual feedback across multiple evaluation criteria simultaneously.

Evaluation and Benchmarking Agent and Tool Ecosystem MGDA Multi-Task Learning LLM-as-a-Judge +4 more