There’s a new player in the generative AI scene, though its name has long been a formidable one in the tech industry.
Amazon Web Services (AWS) recently introduced a new service that offers a number of foundation models that will enable firms to create and customize generative AI applications for a variety of uses.
Known as Amazon Bedrock, the service offers foundation models developed by AI21 Labs, Amazon, Anthropic, and Stability AI, all of which may be accessed via API.
Fewer Data, More Efficiency
According to AWS vice-president for data and machine learning Swami Sivasubramanian, Amazon Bedrock is an excellent way to find the right foundation model to use depending on specific cases and to adjust it accordingly with the relevant enterprise data sets prior to integration and deployment into a number of applications.
As Sivasubramanian puts it, customers can simply point Bedrock at several labeled examples in Amazon S3 and the service can customize the model for a specific task without needing to annotate a large amount of data beforehand.
Among the Amazon foundation models offered on Bedrock are the Titan FMs: two new LLMs trained on a vast range of textual data which can be used to generate human-sounding responses.
The first of these LLMs is a generative model geared towards classification, information extraction, open-ended Q&As, summarization, and text generation. The second translates text input in the form of words, phrases, or larger tracts of text into embeddings or numerical representations containing the semantic meaning of the text.
It should be noted that this second LLM does not generate text, but it can augment a number of applications, including search and personalization since it compares embeddings to produce more relevant and in-context responses.
Amazon representatives add that these foundation models were specifically engineered to detect and remove any harmful content in any enterprise data fed into it for customization. Consequently, these models can also remove harmful content from the final output.
Aside from the Titan foundation models, the new service also includes AI21 Labs’ Jurassic-2 collection of multilingual LLMs which were programmed to obey natural language instructions to generate text in several European languages, including Dutch, French, German, Italian, Portuguese, and Spanish.
Anthropic, on the other hand, brings in Claude, an LLM that performs numerous conversational and text-processing tasks. This model is based on its developers’ longstanding research on training honest and responsible AI systems.