AWS Bedrock Approach vs. Google & OpenAI
AWS Bedrock is the latest GenAI offering on AWS. A successor of its private beta launched in April, Bedrock opens doors to using foundation models (FMs) of leading AI corporations through a single API. Sparing developers the headache of deploying and managing those models themselves.
But it also comes months after Google and OpenAI released their offering. How does it compare? Which one should I choose? And is there a difference at all?
A Plethora of Models to Choose From, but how?
One thing AWS is doing exceptionally well compared to Google. AWS offers open-source LLMs behind the same API interface. I really like this approach and hope we see more models in the near future.
The list of available foundation models on AWS is a great starting point, with offerings ranging from Anthropic’s variants to Amazon’s own Titan series. Moreover, with Meta’s Llama 2 on the horizon, developers have a rich palette of models to select from, each tailored for specific use cases.
It truly feels like a unified experience with Amazon Bedrock.
Google only offers its own PaLM models as fully managed serverless API. They also offer open-source models with their Model Garden and easy deployment, but still, you need to deploy them and always have the models up and running. While Google is currently offering more open-source models, the hosting experience is, therefore, different.
OpenAI obviously follows a different business approach and offers no other models apart from their own (which makes sense).
Interacting with UIs
Getting started with Bedrock is a breeze. Once models are enabled through the console, developers can experiment with them in real time via the Chat Playground. Here, one can tweak parameters like Temperature, Top P, and Top K to influence the response’s character and see the model’s capabilities firsthand.
The UI is a great starting point for drafting and experimenting with your prompt. The AWS UI is very similar to Google and OpenAI. I am missing the possibility to save and manage different prompts.