Nous Research has unveiled Arete-27B, a fine-tuned version of the Qwen3.8-27B model, which aims to address concerns over biased outputs. In a comprehensive evaluation using 803 prompts, the new model demonstrated a remarkable reduction in propaganda answers, decreasing from 22% to just 0.9%.
Despite this significant improvement in bias reduction, the performance on the Hermes Bench remained stable at 65.60%. This development is crucial as it highlights Nous Research’s commitment to creating AI models that prioritize ethical considerations and accuracy in information dissemination.
The Arete-27B model is released under the Apache 2.0 license, making it accessible for further research and development in the AI community. This move not only enhances transparency but also encourages collaboration among developers and researchers to build upon this foundation.
Compiled automatically by the Tech AI Newsdesk from public AI-news sources and summarised in our own words.
Frequently asked questions
What is the Arete-27B model?
Arete-27B is a fine-tuned version of the Qwen3.8-27B model aimed at reducing biased outputs.
How much did propaganda answers decrease in Arete-27B?
The Arete-27B model demonstrated a reduction in propaganda answers from 22% to just 0.9%.
What license is the Arete-27B model released under?
The Arete-27B model is released under the Apache 2.0 license, promoting accessibility for further research.
