# Language model evaluation

> Asking better questions about model quality.

Canonical page: https://rajakarthik.com/work/foundation-models/

## Overview

A single benchmark score never tells the whole story. Evaluation also means examining consistency, instruction following, and the ways models fail on ordinary tasks.

I am interested in test design, qualitative error analysis, and comparing model behavior across different languages and task types.

## Topics I explore

- Instruction-following and response consistency
- Multilingual evaluation principles
- Error analysis and transparent reporting
- Reproducibility and careful interpretation

An evaluation is useful when it helps explain a failure, not just rank a score.