AWS doubles down on custom LLMs with features meant to simplify model creation

In a bold move to empower businesses, Amazon Web Services (AWS) has introduced new tools aimed at simplifying the creation of custom large language models (LLMs). This announcement closely follows the launch of Nova Forge, a service designed for training personalized Nova AI models, and was made during the AWS re:Invent conference on Wednesday. Among the highlights are fresh capabilities within Amazon Bedrock and Amazon SageMaker AI, which are tailored to streamline the process of building and fine-tuning LLMs for developers. One of the standout features is the introduction of serverless model customization in SageMaker. This innovation allows developers to focus on model creation without the complexities of managing compute resources or infrastructure, according to Ankur Mehrotra, general manager of AI platforms at AWS, in an interview with TechCrunch. Developers can engage with these serverless capabilities through a user-friendly point-and-click interface or via an agent-led experience that accepts natural language prompts. The agent-led feature is currently in preview. "If you’re a healthcare customer needing a model that comprehends specific medical terms, you can direct SageMaker AI with your labeled data, choose a technique, and SageMaker will efficiently fine-tune the model," Mehrotra explained. This customization feature extends to Amazon’s own Nova models, as well as select open-source models like DeepSeek and Meta's Llama. Additionally, AWS has rolled out Reinforcement Fine-Tuning in Bedrock, enabling developers to select either a reward function or a pre-defined workflow, allowing for an automated model customization process from start to finish. At this year's conference, AWS appears to be placing significant emphasis on frontier LLMs—the pinnacle of AI technology—and model customization. During Tuesday's keynote by AWS CEO Matt Garman, the company revealed Nova Forge, a service that builds custom Nova models for enterprise clients at an annual cost of $100,000. As Mehrotra noted, many customers are inquiring, "If my competitor has access to the same model, how do I stand out?" He emphasized that the solution lies in developing unique, optimized models tailored to individual business needs. Despite AWS's current challenges in establishing a robust user base for its AI models, as highlighted by a July survey from Menlo Ventures showing a preference for competitors like Anthropic and OpenAI, the newly introduced customization and fine-tuning capabilities could provide AWS with a significant edge in the competitive landscape of AI development.