September 2024 Summaries

Fine-tuning for Claude 3 Haiku, a fast and cost-effective model, is now available in Amazon Bedrock, allowing developers to customize the model with their own data to increase performance for specialized tasks. This enables enhanced domain expertise, cost savings, simplified deployment, and secure model integration. Fine-tuning has been tested on various use cases, including classification, structured outputs, brand voice consistency, and customer support, demonstrating improved performance in financial data analysis and question-answering tasks. SK Telecom, a telecommunications company, has seen significant improvements in customer feedback, key performance indicators, and response quality by training a custom Claude model. To get started, developers can log in to the Amazon Bedrock console and explore additional resources for more information.

Contextual Retrieval is a method that significantly improves the retrieval step in Retrieval-Augmented Generation (RAG). It uses two sub-techniques, Contextual Embeddings and Contextual BM25. This method can reduce the number of failed retrievals by 49% and, when combined with reranking, by 67%. These represent significant improvements in retrieval accuracy, which directly translates to better performance in downstream tasks. Contextual Retrieval is uniquely possible at low cost with Claude, thanks to its special prompt caching feature.

The Anthropic API Console introduces Workspaces to help developers manage multiple Claude deployments efficiently. Workspaces are unique environments that allow for organizing resources, streamlining access controls, and setting custom spend and rate limits on a granular level. This feature is beneficial for developers using Claude across different environments and use cases, as it provides an abstraction layer for overall organization and individual API keys. Key features include setting granular spend limits, grouping related resources, managing rate limits independently, streamlining access control, and monitoring usage efficiently. Workspaces are now available to all Anthropic API users in the Console.

The new Claude Enterprise plan is designed to help organizations securely collaborate using internal knowledge. It offers an expanded context window, increased usage capacity, and a native GitHub integration for working on entire codebases with Claude. Additionally, the plan includes enterprise-grade security features such as SSO, role-based permissions, audit logs, and SCIM. The Enterprise plan enables teams to scale expertise across projects, decisions, and teams while protecting sensitive data. Early customers have used Claude for Work in various tasks, including brainstorming, streamlining internal processes, content creation, and writing code.

Salesforce has partnered with Anthropic to enhance its Einstein capabilities by integrating Claude, a large language model (LLM). The collaboration allows Salesforce customers to use Claude models for AI-powered business applications and experiences built with Einstein 1 Studio. Claude is designed for enterprise-grade security and accuracy, offering near-human voice and tone and less hallucination than other LLMs. This partnership enables users to tailor their CRM applications according to their unique requirements by integrating their own LLMs. The integration of Claude into Einstein 1 Studio offers various use cases across industries such as sales, marketing, customer service, healthcare, financial services, legal, technology, media, and entertainment.