Understanding the Effects of RLHF on LLM Generalisation and Diversity

By Robert Kirk, Ishita Mediratta, Christoforos Nalmpantis, Jelena Luketina, Eric Hambro, Edward Grefenstette, Roberta Raileanu. International Conference on Learning Representations, 2023

411 citations recorded by the source corpus.

Large language models (LLMs) fine-tuned with reinforcement learning from human feedback (RLHF) have been used in some of the most widely deployed AI models to date, such as OpenAI's ChatGPT or Anthropic's Claude. While there has been significant work developing these methods, our understanding of the benefits and downsides of each stage in RLHF is still limited. To fill this gap, we present an extensive analysis of how each stage of the process (i.e. supervised fine-tuning (SFT), reward modelling, and RLHF) affects two key properties: out-of-distribution (OOD) generalisation and output diversity. OOD generalisation is crucial given the wide range of real-world scenarios in which these models are being used, while output diversity refers to the model's ability to generate varied outputs and is important for a variety of use cases. We perform our analysis across two base models on both summarisation and instruction following tasks, the latter being highly relevant for current LLM use cases. We find that RLHF generalises better than SFT to new inputs, particularly as the distribution shift between train and test becomes larger. However, RLHF significantly reduces output diversity compared to SFT across a variety of measures, implying a tradeoff in current LLM fine-tuning methods between generalisation and diversity. Our results provide guidance on which fine-tuning method should be used depending on the application, and show that more research is needed to improve the tradeoff between generalisation and diversity.

Subfields: Artificial Intelligence

View Paper: https://www.semanticscholar.org/paper/cb3968152f7d93f53d24b00279a90d5071ddc85a

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

Citations in the corpus

ORPO: Monolithic Preference Optimization without Reference Model (2024, methodology): “We expand the concept of per-input and across-input diversity introduced in Kirk et al. (2024) by using Gemini-Pro (Gemini Team et al., 2023) as an embedding model, which is suitable for assessing the diversity of instruction-following language models by encoding a maximum of 2048 tokens.”
Towards Understanding Sycophancy in Language Models (2023, neutral): “The algorithm used to optimize the PM also affects properties of the policy, such as diversity and generalization (Kirk et al., 2023).”
Are We Measuring Strategy or Phrasing? The Gap Between Surface- and Approach-Level Diversity in LLM Math Reasoning (2026, unknown): “…learning with verifiable rewards (RLVR), a dominant paradigm for post-training language models, reduces the diversity of a trained policy (Kirk et al., 2024; He et al., 2025; Yue et al., 2025; Wu and Choi, 2025), prompting diversity-aware RLVR methods that attempt to preserve diversity…”