During the AWS re:Invent generative AI keynote, Amazon announced Bedrock enactment for Claude 2.1 and Llama 2 70B and more.
After the AWS announcements yesterday astir the Amazon Q chatbot for endeavor and almighty caller chips for AI workloads, Vice President of Databases, Analytics and Machine Learning astatine AWS Swami Sivasubramanian took the signifier astatine the AWS re:Invent league successful Las Vegas connected Nov. 29 to dive deeper into AWS AI offerings. Sivasubramanian announced caller generative AI models coming to Amazon Bedrock, multimodal searching disposable for Amazon Titan successful Amazon Bedrock and galore different caller endeavor bundle features and tools related to utilizing generative AI for work.
Jump to:
- Amazon Titan tin present tally searches based connected substance and images
- Titan Text Lite and Titan Text Express added to Amazon Bedrock
- Claude 2.1 and Llama 2 70B present hosted connected Amazon Bedrock
- Additional Amazon Q usage cases announced
- Training and exemplary valuation tools added to Amazon SageMaker
- Vector capabilities and information absorption tools added to galore AWS services
Amazon Titan tin present tally searches based connected substance and images
Amazon Titan Multimodal embeddings are present successful wide availability successful Amazon Bedrock, the AWS instrumentality for gathering and scaling AI applications. Multimodal embeddings let organizations to physique applications that fto users hunt utilizing substance and images for richer hunt and proposal options, said Sivasubramanian.
“They (AWS customers) privation to alteration their customers to hunt for furnishings utilizing a phrase, representation oregon adjacent both,” said Sivasubramanian. “They could usage instructions similar ‘show maine what works good with my sofa’.”
SEE: Are AWS oregon Google Cloud close for your business? (TechRepublic)
Titan Text Lite and Titan Text Express added to Amazon Bedrock
Titan Text Lite and Titan Text Express are present mostly disposable successful Amazon Bedrock to assistance optimize for accuracy, show and cost, depending connected their usage cases. Titan Text Lite is simply a precise tiny exemplary for substance and tin beryllium fine-tuned. Titan Text Express is simply a exemplary that tin bash a wider scope of text-based generative AI tasks, specified arsenic conversational chat and open-ended questions.
Titan Image Generator (Figure A) is present disposable successful nationalist preview successful the U.S. It tin beryllium utilized to make images utilizing earthy connection prompts. Organizations tin customize images with proprietary information to lucifer their manufacture and brand. Images volition beryllium invisibly watermarked by default to assistance debar disinformation.
Figure A
An representation created by Titan Image Generator. Image: AWSClaude 2.1 and Llama 2 70B present hosted connected Amazon Bedrock
Amazon Bedrock volition present enactment Anthropic’s Claude 2.1 for users successful the U.S. This mentation of the Claude generative AI offers advancements successful a 20,000 discourse window, improved accuracy, 50% less hallucinations adjacent during adversarial punctual attacks and 2 times simplification successful mendacious statements successful open-ended conversations compared to Claude 2. Tool usage for relation calling and workflow orchestration successful Claude 2.1 are disposable successful beta for prime aboriginal entree partners.
Meta’s Llama 2 70B, a nationalist ample connection exemplary fine-tuned for chat-based usage cases and large-scale tasks, is disposable contiguous successful Amazon Bedrock.
Claude assistance disposable successful AWS Generative AI Innovation Center
The AWS Generative AI Innovation Center volition grow aboriginal successful 2024 with a customized exemplary programme for Anthropic Claude. The AWS Generative AI Innovation Center is designed to assistance radical enactment with AWS’ squad of experts to customize Claude needs for one’s ain proprietary concern data.
Additional Amazon Q usage cases announced
Sivasubramanian announced a preview of Amazon Q, the AWS earthy connection chatbot, successful Amazon Redshift, which tin supply assistance with penning SQL. Amazon Redshift with Amazon Q lets developers inquire earthy connection questions, which the AI translates into a SQL query. Then, they tin tally that query and set it arsenic necessary.
Plus, Amazon Q for information integration pipelines is present disposable connected the serverless computing level AWS Glue for gathering information integration jobs successful earthy language.
Training and exemplary valuation tools added to Amazon SageMaker
Sivasubramanian announced the wide availability of SageMaker HyperPod, a caller distributed generative AI grooming capableness to trim exemplary grooming clip up to 40%. SageMaker HyperPod tin bid generative AI models connected its ain for weeks oregon months, automating the tasks of splitting information into chunks and loading that information onto idiosyncratic chips successful a grooming cluster. SageMaker HyperPod includes SageMaker’s distributed grooming pods, managed checkpoints for optimization, the quality to observe and reroute astir hardware failures. Other caller SageMaker features see SageMaker inference for faster optimization and a caller idiosyncratic acquisition successful SageMaker Studio.
Amazon SageMaker and Bedrock present person Model Evaluation, which lets customers measure antithetic instauration models to find which is the champion for their usage case. Model Evaluation is disposable successful preview.
Vector capabilities and information absorption tools added to galore AWS services
Sivasubramanian announced much caller tools astir vectors and information absorption that are suitable for a assortment of endeavor usage cases, including generative AI.
- Vector Engine for OpenSearch Serverless is present mostly available.
- Vector capabilities are coming to Amazon DocumentDB and Amazon DynamoDB (out present successful each regions wherever Amazon DocumentDB is available) and Amazon MemoryDB for Redis (now successful preview).
- Amazon Neptune Analytics, an analytics database motor for Amazon Neptune oregon Amazon S3, is disposable contiguous successful certain regions.
- Amazon OpenSearch work zero-ETL integration with Amazon S3.
- AWS Clean Rooms ML, which lets organizations stock instrumentality learning models with partners without sharing their underlying data.
“While gen AI inactive needs a beardown foundation, we tin besides usage this exertion to code immoderate of the large challenges successful information management, similar making information easier to use, making it much intuitive and making information much valuable,” Sivasubramanian said.
Note: TechRepublic is covering AWS re:Invent virtually.
English (US) ·