ME News reports that on November 6 (UTC+8), Reflection published a blog post detailing its secure training and evaluation methods for its open-source weighted model, Beam, and launched the open-source security research initiative Project Iris. Beam rejected 89% of harmful requests in the Iris Safety Evaluations dialogue task—the highest among all tested open-source weighted models—while complying with 82% of safety-related harmless requests. After scaling the safety-aligned teacher model from 100B to 501B parameters, the rate of over-rejection decreased by 48.7%.
🔗 Read the original article
via AIHOT · https://aihot.news/items/mpv0jfejdzdm2336jgtyj9tde (Source: AiHot)
Reflection Details: Safe Training and Evaluation Methods for Open-Source Model Beam
KuCoinFlashShare
Summary
On-chain news from Reflection introduces new safety measures for its open-source model, Beam, announced on November 6 (UTC+8). The blog highlights Beam’s 89% rejection rate for harmful requests in Iris Safety Evaluations, with 82% compliance on harmless safety requests. The risk of security breaches was reduced as over-rejection decreased by 48.7% following the expansion of the safety-aligned teacher model to 501B parameters. Project Iris now supports open-source security research.
Source:Show original
Disclaimer: The information on this page may have been obtained from third parties and does not necessarily reflect the views or opinions of KuCoin. This content is provided for general informational purposes only, without any representation or warranty of any kind, nor shall it be construed as financial or investment advice. KuCoin shall not be liable for any errors or omissions, or for any outcomes resulting from the use of this information.
Investments in digital assets can be risky. Please carefully evaluate the risks of a product and your risk tolerance based on your own financial circumstances. For more information, please refer to our Terms of Use and Risk Disclosure.