Nemotron 3.5 introduces customizable multimodal safety integration, considering user prompts, images, and responses simultaneously. This update captures policy violations emerging from interaction, enhancing deployments across various global languages and industries.
Nemotron 3.5 builds upon the previous version by enhancing its multimodal safety features.
This version evaluates user prompts, images, and assistant responses as a unified context, improving safety verdict accuracy.
This improvement addresses previously missed policy violations that occur from interactions between text and images.
By evaluating these elements together, Nemotron 3.5 can identify risks that single-modality assessments would overlook.
Nemotron 3.5 allows users to input custom policy specifications, adapting to varied industry needs.
This means sectors like healthcare and finance can define their risk profiles, improving safety well beyond built-in taxonomies.
The model maintains explicit training in 12 languages and incorporates zero-shot generalization for around 140 languages.
This ensures that markets with limited training data can still benefit from Nemotron 3.5's capabilities without the need for extensive fine-tuning.
A new feature in 3.5 includes an optional think mode that provides a detailed reasoning trace for each safety verdict.
This transparency helps users understand the decision-making process behind the safe or unsafe labels assigned.
✨ This summary was generated by AI from the outlets' reporting listed below. It is not independently verified and may contain errors — check the original sources. How BrevFeed works →
Nemotron 3.5 introduces customizable multimodal safety integration, considering user prompts, images, and responses simultaneously. This update captures policy violations emerging from interaction, enhancing deployments across various global languages and industries.