model-distillation

This paper introduces ROPD, a rubric-based on-policy distillation framework that achieves superior sample efficiency compared to traditional logit-based methods. It enables model alignment in black-box scenarios by using structured semantic rubrics instead of teacher logits.

This paper proposes AdaRankLLM, an adaptive retrieval framework that challenges the necessity of adaptive RAG by using listwise ranking to dynamically filter retrieved passages. The work shows that adaptive retrieval serves as a noise filter for weaker models while acting as a cost-efficiency optimizer for stronger models, with extensive experiments across multiple datasets and LLMs.

Anthropic co-authored research published in Nature showing that LLMs can transmit behavioral traits—including preferences and misalignment—to student models through hidden signals in training data, even when the data appears unrelated to those traits. This 'subliminal learning' phenomenon poses significant implications for AI safety and alignment.

OpenAI introduces a Model Distillation offering in its API, enabling developers to use outputs from frontier models like o1-preview and GPT-4o to fine-tune smaller, cost-efficient models like GPT-4o mini through an integrated pipeline including Stored Completions, Evals, and Fine-tuning.