OpenAI fine-tunes 774M-parameter GPT-2 with human preferences
Original title:Fine-tuning GPT-2 from human preferences
AI summary
OpenAI fine-tuned a 774M-parameter GPT-2 with human feedback, matching external labelers' preferences on multiple tasks. Summarization used 60,000 human labels, while simpler tasks such as style continuation needed only 5,000; labelers preferred summaries that copied sentences verbatim from the source, leading the model to learn copying. OpenAI said the effort aims to bring safety techniques closer to the general task of machine-human dialogue.
Key Takeaways
- OpenAI fine-tuned a 774M-parameter GPT-2 with human feedback,matching external labelers'…
- Summarization used 60,000 human labels,while simpler tasks such as style continuation needed only…
- labelers preferred summaries that copied sentences verbatim from the source,leading the model to…
- OpenAI said the effort aims to bring safety techniques closer to the general task of machine-human…
Quick Facts
- Published
- Updated
- Source
- OpenAI:官网动态(RSS · 排除企业/客户案例)
- Category
- Papers & Research
- AI score
- 65
原文给出人类偏好微调 GPT-2 的任务结果与标注成本,并暴露标注偏好与作者意图不一致的问题。
Related entities
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: OpenAI fine-tunes 774M-parameter GPT-2 with human preferences. September 19, 2019. https://new.nat6.net/en/news/mbl0jky4zuwkwo6nc6declud6/