AI Insight
CRAFT is a new evaluation method that diagnoses specific capability weaknesses in large language models by analyzing grading rubrics rather than just identifying poor performance areas. The method clusters rubric criteria into a hierarchical capability tree, identifies weak nodes, and generates targeted fine-tuning data to address those specific deficiencies. Testing on four open-source models across finance and legal domains showed CRAFT produced stronger performance improvements compared to baseline approaches, with the best results in finance across all models and in legal for three of four models.
Why it matters
This approach provides a more precise diagnostic framework for improving AI systems by identifying not just where models fail but why they fail at the capability level. The method offers a systematic way to generate targeted training data that addresses specific weaknesses, potentially making model improvement more efficient and cost-effective than untargeted fine-tuning approaches.
Understand the Science
arXiv:2607.16122v1 Announce Type: cross
Abstract: Evaluations should do more than measure a models current performance. They should tell us what to fix for the next model iteration and provide a way to generate targeted post training data. Most evaluation pipelines identify weak examples, topics, or categories, but they leave the underlying capability failure implicit: they say where a model fails, not why. We introduce CRAFT, a method that converts any rubric based evaluation dataset into a model specific diagnosis of weak capabilities. CRAFT treats each grading criterion as a capability probe: it extracts a capability description from every prompt rubric pair, clusters these descriptions into a hierarchical capability tree, scores the target model at every node, and selects low performing nodes dynamically across tree levels, at the granularity where each failure is clearest. The selected weak capabilities then direct the generation of targeted supervised finetuning data. Holding the data generation, finetuning, and evaluation setup fixed, we compare CRAFT against prompt level EvalTree clustering and untargeted random generation on four open source models, two professional domains (finance and legal), and 13 held out benchmarks disjoint from the diagnostic data. CRAFT achieves the strongest finance domain average for all four models under repeated temperature decoding; on legal domain, it is strongest for three of four models and remains within the decoding variance bands of the best baseline on the fourth. Diagnosing weaknesses at the level of rubric criteria, rather than prompts or categories, thus yields both a sharper picture of what a model cannot do and measurably better models after finetuning on that diagnosis.
Source: CRAFT: Clustering Rubrics to Diagnose Weak LLM Capabilities and Generate Targeted Fine-Tuning Data