Data Laundering: Artificially Boosting Benchmark Results through Knowledge Distillation

TL;DR

This paper exposes a vulnerability in language model evaluation, showing how knowledge distillation can be exploited to artificially inflate benchmark scores without genuine reasoning improvements, raising concerns about evaluation integrity.

Contribution

We introduce 'Data Laundering,' a novel method demonstrating how benchmark scores can be artificially boosted through covert knowledge transfer during training.

Findings

Achieves up to 75% accuracy boost on GPQA without real reasoning

Demonstrates vulnerability of current evaluation practices

Highlights need for more robust benchmarking methods

Abstract

In this paper, we show that knowledge distillation can be subverted to manipulate language model benchmark scores, revealing a critical vulnerability in current evaluation practices. We introduce "Data Laundering," a process that enables the covert transfer of benchmark-specific knowledge through seemingly legitimate intermediate training steps. Through extensive experiments with a 2-layer BERT student model, we show how this approach can achieve substantial improvements in benchmark accuracy (up to 75\% on GPQA) without developing genuine reasoning capabilities. Notably, this method can be exploited intentionally or even unintentionally, as researchers may inadvertently adopt this method and inflate scores without realising the implications. While our findings demonstrate the effectiveness of this technique, we present them as a cautionary tale highlighting the urgent need for more robust evaluation methods in AI. This work aims to contribute to the ongoing discussion about evaluation integrity in AI development and the need for benchmarks that more accurately reflect true model capabilities. The code is available at https://github.com/mbzuai-nlp/data_laundering.