Bias Amplification in Language Model Evolution: An Iterated Learning Perspective

Authors: Yi Ren, Shangmin Guo, Linlu Qiu, Bailin Wang, Danica J. Sutherland

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental This paper outlines key characteristics of agents behavior in the Bayesian-IL framework, including predictions that are supported by experimental verification with various LLMs.
Researcher Affiliation Academia Yi Ren UBC renyi.joshua@gmail.com Shangmin Guo University of Edinburgh s.guo@ed.ac.uk Linlu Qiu MIT linluqiu@mit.edu Bailin Wang MIT bailin.wang28@gmail.com Danica J. Sutherland UBC & Amii dsuth@cs.ubc.ca
Pseudocode No The paper describes algorithms and procedures conceptually but does not include a formal pseudocode block or algorithm listing.
Open Source Code Yes The code for experiments is available at https://github.com/Joshua-Ren/i ICL.
Open Datasets Yes We finetune a pretrained llama-2-7B model (Touvron et al. 2023) using Antropic-HH dataset (Bai et al. 2022)
Dataset Splits No The paper describes initial data (`d0`) and data pools (`Dpool`) used in experiments, but it does not specify explicit training, validation, or test dataset splits (e.g., percentages, sample counts, or predefined split citations) needed for reproducibility.
Hardware Specification No The paper mentions the use of LLMs (GPT3.5, GPT4, Claude3-haiku, Mixtral-8x7b) for experiments, but it does not specify the underlying hardware (e.g., GPU/CPU models, memory, or specific cloud instances) used for running these experiments.
Software Dependencies No The paper mentions using specific LLMs (GPT3.5, GPT4, Claude3-haiku, Mixtral-8x7b) but does not provide details on specific software dependencies, libraries, or frameworks with version numbers (e.g., Python, PyTorch, TensorFlow) for reproducibility.
Experiment Setup Yes The temperature is 0.1 and the probability feedback is enabled.