Amazon Web Services (AWS) this week unveiled Nova, its latest suite of multi-modal AI models. We jumped in to try them out using iMerit’s Ango Hub software. Our seamless integration with these models has allowed us to get up and running to conduct initial testing across various use cases. Let’s dive into our findings with the three new AWS Nova variants.
The Nova Family: A Quick Overview
Nova Micro
The most streamlined of the bunch, Nova Micro specializes in text-only processing. Our testing revealed impressive capabilities in language translation, particularly with complex text passages from English to Spanish. The translations maintained contextual accuracy while preserving technical details and nuanced meaning.
Nova Lite
Nova Lite introduces multi-modal capabilities, making it particularly effective for visual analysis tasks. We extensively tested it with traffic sign datasets, and the results were noteworthy. The model demonstrated strong performance in:
- Identifying traffic signs across different contexts
- Understanding and interpreting prohibition signs
Nova Pro
Building on Lite’s capabilities, Nova Pro takes multi-modal analysis a step further. One of its standout features is the ability to detect and identify the language used in signs and visual content. This makes it particularly valuable for international applications and multi-lingual environments.
Real-World Testing Examples
Micro Model Translation Capabilities:
Our tests with Nova Micro’s translation features yielded impressive results across various content types:
1. Political News: Accurately translated complex content about the Kyrgyz Republic’s election procedures, maintaining technical accuracy and cultural context.
Figure: Political Bulletins Translation on Ango
Translated Text: “ La República Kirguisa, un antiguo estado de la ex Unión Soviética, está utilizando tinta invisible y lectores ultravioleta en sus elecciones para evitar la votación múltiple. Esta iniciativa, impulsada por el presidente Askar Akaev, tiene como objetivo mantener la reputación democrática del país. El uso de la tinta ha generado reacciones mixtas, algunos la ven como un paso hacia elecciones más transparentes, mientras que otros la comparan con marcar al ganado. A pesar de las controversias y rumores, la medida es parte de un esfuerzo más amplio para garantizar elecciones libres y justas, con el apoyo de entidades internacionales como el gobierno de EE. UU., la Embajada Alemana y la Fundación Soros.”
2. Technical Bulletins: Successfully handled Microsoft security update announcements, preserving technical terminology and maintaining clarity.
Figure: Technical Bulletins Translation on Ango
Translated Text: “Microsoft ha emitido una advertencia a los usuarios de PC para que actualicen sus sistemas con los últimos parches de seguridad para abordar vulnerabilidades críticas en los programas de Windows. En su boletín de seguridad mensual, Microsoft identificó ocho vulnerabilidades “críticas” que podrían exponer las PC a ataques si no se corrigen. Estas vulnerabilidades afectan a varios programas de Windows, incluyendo Internet Explorer, el reproductor multimedia y la mensajería instantánea. Además, se publicaron cuatro correcciones importantes adicionales, aunque se consideraron menos críticas. Si los usuarios no actualizan sus sistemas, podrían volverse susceptibles a virus o ataques maliciosos que exploten estas vulnerabilidades. Muchas de estas fallas podrían ser utilizadas por ciberdelincuentes para controlar remotamente las computadoras, instalar programas, alterar, eliminar o ver datos.
Uno de los parches críticos que Microsoft ha puesto a disposición corrige algunas fallas de Internet Explorer, conocidas por la empresa pero que aún no han sido explotadas en ataques. Otro parche corrige una vulnerabilidad en Media Player, Windows Messenger y MSN Messenger que podría ser explotada a través de archivos .png para tomar el control de máquinas desprotegidas. Microsoft anuncia regularmente vulnerabilidades de software, siendo las más graves las clasifica”
3. Sports Technology: Effectively translated detailed content about BBC Sport’s Piero analysis tool, keeping specialized terms and concepts intact.
Figure: Sport technology Text Translation on Ango
Translated Text: “BBC Sport presenta su nueva herramienta de análisis Piero en el partido de rugby entre Gales e Inglaterra el sábado. Piero, nombrado en honor al pintor italiano Piero della Francesco, está diseñado para crear un estadio virtual donde los jugadores pueden ser rastreados desde casi cualquier ángulo. Esta tecnología proporciona a los espectadores información detallada sobre la trayectoria del balón y los movimientos de los jugadores, mejorando el análisis durante las transmisiones deportivas. Desarrollado por BBC Research and Development, Piero utiliza datos telemétricos de cámaras fijas para renderizar gráficos 3D, ofreciendo reproducciones casi en tiempo real y análisis en profundidad. BBC Sport busca mantener a las audiencias comprometidas con respuestas más detalladas y definitivas sobre momentos clave del juego. La herramienta también atrae el interés de empresas de rendimiento deportivo y podría venderse a terceros para diferentes aplicaciones más allá de las transmisiones deportivas.”
Nova Lite and Pro’s Traffic Sign Analysis
The multi-modal capabilities of Nova Lite and Pro excelled in our traffic sign analysis tests. The models demonstrated:
- High accuracy in sign classification
- Strong contextual understanding
- Language detection capabilities (Nova Pro)
Example Images from the analysis:
Figure: Traffic Signs Detection
Figure: Traffic Signs Detection
Figure: Traffic Signs Detection by Nova Pro
Figure: Traffic Signs Detection by Nova Pro
Our testing of Nova Lite with traffic sign datasets revealed both capabilities and limitations. Notably, we uncovered some consistency issues in sign interpretation. In one significant example, when presented with identical crossmark symbols in different images, the model provided contradictory interpretations:
- For one instance, it identified the symbol as indicating “parking is prohibited in this area”
- For the same symbol in another context, it interpreted it as “Vehicles cannot cross the intersection”
“parking is prohibited in this area”
“Vehicles cannot cross the intersection”
Implementation on Ango Hub
Setup and Integration
The integration of AWS Nova models into Ango Hub is straightforward. The platform provides a simple plugin architecture that allows for quick model deployment. As shown in our implementation, the process involves:
- Plugin Configuration
- Models are available directly through the Ango Hub interface
- Simple dropdown selection for different Nova models (Pro, Lite, Micro)
- Integrated configuration panel for model settings and parameters
Figure: AWS Nova workflow on Ango
Figure: Loading Assets onto Ango
Key Takeaways
The AWS Nova model family shows promise in multi-modal AI capabilities, and our testing has revealed several important insights:
Model Capabilities
- Nova Micro demonstrates strong performance in pure text processing and translation
- Nova Lite shows potential in multi-modal analysis but needs improvement in consistency, particularly for similar visual elements
- Nova Pro adds language detection capabilities to the multi-modal framework
Implementation Benefits on Ango Hub
- Seamless integration of all three Nova models into the Ango Hub platform
- Zero additional configuration required – models work out of the box
- Easy switching between different Nova models for comparative testing
- Integrated workflow support for both text and image processing tasks
- Streamlined deployment process with minimal technical overhead
Leveraging Nova and Human-in-the-loop in tandem
One important option is that the Ango Hub workflow can leverage models like Nova as a pre-processing model and then add an expert in the loop to review/audit/correct errors. This results in higher final quality, can catch egregious errors (as we did in the traffic signs) and increase data labeling or data auditing throughput by a significant margin.
Looking Forward
The combination of AWS Nova’s capabilities with iMerit’s Ango Hub platform creates a powerful, user-friendly environment for AI implementation. The platform’s ability to seamlessly integrate these advanced models while maintaining ease of use makes it an attractive option for businesses looking to leverage AWS’s latest AI capabilities without extensive technical overhead.
