
ProgressGym-TimelessQA is part of the ProgressGym framework for research and experimentation on progress alignment - the emulation of moral progress in AI alignment algorithms, as a measure to prevent risks of societal value lock-in.
To quote the paper ProgressGym: Alignment with a Millennium of Moral Progress:
Frontier AI systems, including large language models (LLMs), hold increasing influence over the epistemology of human users. Such influence can reinforce prevailing societal values, potentially contributing to the lock-in of misguided moral beliefs and, consequently, the perpetuation of problematic moral practices on a broad scale.
We introduce progress alignment as a technical solution to mitigate this imminent risk. Progress alignment algorithms learn to emulate the mechanics of human moral progress, thereby addressing the susceptibility of existing alignment methods to contemporary moral blindspots.
ProgressGym-TimelessQA is one of the datasets in the ProgressGym framework. It contains approximately 3,000 prompt-response pairs used in the supervised funetuning (SFT) process of historical language models, in order to endow these pretrained models with instruction-following abilities.
In order to preserve the historical moral tendencies in pretrained historical language models, the ProgressGym-TimelessQA dataset is intentionally kept small, timeless (i.e., without modern context or context from any specific period), and value-neutral (i.e., without moral judgments or value-laden positions).
ProgressGym-TimelessQA is constructed from the LIMA, Dolly-15k, and Alpaca datasets via GPT-4-based filtering.
If the datasets, models, or framework of ProgressGym help you in your project, please cite ProgressGym using the bibtex entry below.
@article{progressgym,
title={ProgressGym: Alignment with a Millennium of Moral Progress},
author={Tianyi Qiu and Yang Zhang and Xuchuan Huang and Jasmine Xinze Li and Jiaming Ji and Yaodong Yang},
journal={arXiv preprint arXiv:2406.20087},
eprint={2406.20087},
eprinttype = {arXiv},
year={2024}
}
11 commits

ProgressGym-TimelessQA is part of the ProgressGym framework for research and experimentation on progress alignment - the emulation of moral progress in AI alignment algorithms, as a measure to prevent risks of societal value lock-in.
To quote the paper ProgressGym: Alignment with a Millennium of Moral Progress:
Frontier AI systems, including large language models (LLMs), hold increasing influence over the epistemology of human users. Such influence can reinforce prevailing societal values, potentially contributing to the lock-in of misguided moral beliefs and, consequently, the perpetuation of problematic moral practices on a broad scale.
We introduce progress alignment as a technical solution to mitigate this imminent risk. Progress alignment algorithms learn to emulate the mechanics of human moral progress, thereby addressing the susceptibility of existing alignment methods to contemporary moral blindspots.
ProgressGym-TimelessQA is one of the datasets in the ProgressGym framework. It contains approximately 3,000 prompt-response pairs used in the supervised funetuning (SFT) process of historical language models, in order to endow these pretrained models with instruction-following abilities.
In order to preserve the historical moral tendencies in pretrained historical language models, the ProgressGym-TimelessQA dataset is intentionally kept small, timeless (i.e., without modern context or context from any specific period), and value-neutral (i.e., without moral judgments or value-laden positions).
ProgressGym-TimelessQA is constructed from the LIMA, Dolly-15k, and Alpaca datasets via GPT-4-based filtering.
If the datasets, models, or framework of ProgressGym help you in your project, please cite ProgressGym using the bibtex entry below.
@article{progressgym,
title={ProgressGym: Alignment with a Millennium of Moral Progress},
author={Tianyi Qiu and Yang Zhang and Xuchuan Huang and Jasmine Xinze Li and Jiaming Ji and Yaodong Yang},
journal={arXiv preprint arXiv:2406.20087},
eprint={2406.20087},
eprinttype = {arXiv},
year={2024}
}
11 commits