AWS recently announced innovations for Amazon Bedrock, its generative AI service. These new features expand the template and resource options available, providing AWS customers with a more complete experience developing custom generative AI applications.
Among the improvements, we highlight the new models provided by renowned companies and also the customized AWS model called Amazon Titan. This unique Amazon Bedrock model was carefully created and pre-trained by AWS based on large, diverse datasets.
Among the new features presented, we have:
Claude 2.1 from Anthropic: which is the latest version of your language model. With an expanded context window to 200 thousand tokens and greater precision, Claude 2.1 allows you to process large documents, such as financial statements, perform summaries and comparisons between documents. Compared to the previous model, Claude 2.1 achieved a significant reduction in false statements, providing more reliable results.
Meta Llama 2 70B: which represents the next generation of Meta language models. Trained with 40% more data than its predecessor, Llama 1, and twice the context length, this 70 billion parameter model is now available on Amazon Bedrock. In addition, we also have the Llama 2 model, with 13 billion parameters.
Amazon Titan Image Generator: is available in the preview phase and promises to help clients in the advertising, e-commerce and entertainment sectors produce images with professional quality. This model allows you to create accurate compositions and reduce the generation of harmful content. Additionally, all images generated by Amazon Titan are given an invisible watermark by default, providing an unobtrusive way to identify AI-generated images.
Amazon Titan Multimodal Embeddings: a model that helps customers drive more accurate multimodal experiences in searches and recommendations. This model converts images and short texts into embeddings, allowing you to understand the semantic meanings and relationships between data stored in the client's vector database.
AWS Vice President of Data and AI Dr. Swami Sivasubramanian says generative AI is poised to be the most transformative technology of our time. It highlights how customers are applying this technology to new opportunities and overcoming business challenges. According to him, Amazon Bedrock is the preferred choice for customers because it offers market-leading models, customization features and enterprise privacy in a fully managed experience.
New advanced comparison and customization features: In the Amazon Bedrock console, which is available in preview, you can choose and compare models based on specific metrics such as relevance, style, and brand identity. Model performance evaluations and reports take into account key criteria.
Knowledge Bases for Amazon Bedrock: Knowledge Bases for Amazon Bedrock lets you customize model responses with contextual data and establish secure connections to proprietary data sources. These knowledge bases are fully managed, allowing customers to pinpoint the location of their data.
Cohere Command, Meta Llama 2, and Amazon Titan adjustable models on Amazon Bedrock: You can now fine-tune models in Amazon Bedrock to improve accuracy. With incremental training, you can get a more accurate and personalized private model, providing relevant answers. Customer data is encrypted during transmission and storage, keeping it secure and private.
Agents for Amazon Bedrock enable generative AI applications to perform multi-step tasks using enterprise systems and data sources: Now available globally, agents automatically analyze requests and break them down into a logical sequence, using the model's reasoning capabilities to determine the information needed. This enables customers to improve the accuracy and development speed of their generative AI applications.
Guardrails for Amazon Bedrock let you implement guardrails on your models based on your application requirements and Responsible AI policies: guardrails drive response to undesirable content. You can set limits for hate speech, sexual language and violence by filtering the content. In 2024, it will be possible to define filters for offensive and personalized words, blocking interactions between users and models.
Generative artificial intelligence is not just a trend, it is becoming increasingly consolidated, so don't waste any more time! Contact Flexa Cloud and start shaping the future with generative AI, let's go on this journey of innovation and transformation together!
