Hey Zapp, can you tell us about Generative AI at Adobe?
Generative AI at Adobe.
Adobe leverages Generative AI across various products and initiatives to enhance the creative process and improve customer experiences. Here are some of the ways Generative AI is used at Adobe:
Creative Tools IntegrationGenerative AI models are integrated into Adobe's creative tools to assist with content creation and editing. This includes applications like Photoshop, Illustrator, and the Substance and Aero products, where Generative AI capabilities help automate tasks and inspire new designs.
Adobe FireflyThis is Adobe's family of Generative AI models, which is built on top of the Sensei platform. Firefly is designed to support creative professionals by providing AI-powered assistance in generating and editing content.
Content Authenticity Initiative (CAI)Adobe uses Generative AI to support the CAI, which aims to provide transparency and trust in digital media by verifying the authenticity and origin of content.
Adobe SenseiThis is Adobe's AI and machine learning technology that powers intelligent features across all Adobe products. It includes capabilities for image analysis, pattern recognition, and language understanding, which are enhanced by Generative AI.
Training Data CurationAdobe's AI Training Data team curates high-quality, ethically sourced datasets to help build and train Generative AI models for different document types, such as images, videos, text, vectors, and 3D content.
Experience PlatformAdobe Experience Platform uses Generative AI to provide personalized experiences to users by leveraging AI-driven insights and recommendations.
Document CloudGenerative AI is used to enhance functionalities in Acrobat and other Document Cloud services, such as summarization, Q&A capabilities, and content transformation.
Adobe StockGenerative AI is applied to improve search and discovery, content quality, and content similarity models, enhancing the user experience for stock content.
Research and DevelopmentAdobe Research labs work closely with product teams to explore and develop new Generative AI models and applications, focusing on creative professional use cases and next-generation content editing paradigms.Adobe's use of Generative AI is deeply integrated into its product ecosystem, aiming to streamline workflows, enhance creativity, and provide more personalized and engaging user experiences.
Can you tell us more about the AI Training Data Team?
AI Training Data Team at Adobe.
The AI Training Data team at Adobe is focused on building and curating high-quality, comprehensive, and ethically sourced AI training datasets.
These datasets are crucial for the development of machine learning models across various document types, including images, videos, text, vectors, and 3D content. The team's efforts are instrumental in supporting Adobe Research teams in creating models that are integrated into Adobe's suite of creative tools.
Key responsibilities and activities of the AI Training Data team include:
- Ensuring the datasets are representative and diverse to minimize biases in the AI models.
- Maintaining ethical standards in data sourcing and handling, adhering to privacy and data protection regulations.
- Collaborating with researchers and engineers to understand the specific data requirements for different AI projects.
- Developing processes for efficient data labeling and annotation to train machine learning models accurately.
- Implementing quality control measures to ensure the reliability and relevance of the training data.
- Facilitating the continuous improvement of AI models by providing feedback loops from model performance back to data curation.
This team plays a pivotal role in the AI ecosystem at Adobe, enabling the development of advanced AI features and functionalities that enhance the capabilities of Adobe's creative and document management software.
Can you share details of the tools and technologies used by this team?
Tools used by the AI Training Data Team at Adobe.
Here are some of the tools and technologies they likely use:
Data Annotation Tools:
To label and annotate data, which could include proprietary Adobe tools or third-party platforms.
Distributed Computing Frameworks:
Such as Apache Spark for processing large volumes of data.
Cloud Platforms:
Utilizing services from AWS, Azure, or Adobe's internal cloud infrastructure for scalable storage and computing resources.
Machine Learning Frameworks:
Like PyTorch and TensorFlow for developing and training AI models.
Data Storage and Databases:
Using systems like Hadoop HDFS, Amazon S3, or other data lakes for storing datasets.
Containerization:
Employing Docker or Kubernetes for creating reproducible environments and scaling training processes.
Workflow Management Tools:
Such as Apache Airflow for orchestrating complex data pipelines.
GPU Clusters:
For efficient distributed model training, leveraging high-performance computing resources.
AI Model Conversion and Quantization Tools:
To optimize models for cloud and on-device execution.
These tools and technologies enable the team to handle various document types, including images, videos, text, vectors, and 3D content, and support the distributed training and at-scale inference required for Adobe's AI initiatives.
