Amazon CTO Werner Vogels, a man of immense wealth who bought up a small Airbnb in the center of Amsterdam where he lived during the COVID-19 pandemic, Are you spending your days? At first glance, it looks like you're building an AI-powered meeting summary app. Go figure.
In this week's post on Vogels' personal blog, he details Distill, an open source app he built with his “OCTO'' (Office of the CTO) team to transcribe and summarize conference calls. . Distill takes audio recordings of meetings (in formats like MP3, FLAC, WAV, etc.), analyzes them, and generates summaries along with a list of to-do items. Optionally, you can spit out summaries and lists to platforms like his Slack via custom integrations.
Example summary from Vogel's Distill conference summarizer powered by Amazon technology. Image credit: Distill
As you would expect from an app from Amazon's CTO, Distill clearly relies on Amazon's paid products and services to do the computationally heavy lifting. AWS Transcribe performs Distill transcription. Amazon S3 provides storage for conference audio files. Bedrock, Amazon's generative AI development suite, then handles the summarization.
But why create a meeting summarizer when there are countless tools out there that can serve the purpose? Well, you have to imagine that's what Vogels thought. He seems to have a large amount of resources at his disposal and plenty of spare time for his hobby programming project. According to his blog, he is already working on porting his Distill codebase from Python to Rust. (It would be a great job to become a CTO.)
One of the unique things about Distill is that you can choose an AI model to perform your meeting summaries. The default is the Sonnet, a mid-range model from Anthropic's Claude 3 family. (Amazon's large stake in Anthropic may have had something to do with its design decisions.) But other models hosted on Bedrock, such as Meta's Llama 3 and models from AI startups Mistral, AI21 Labs, and Cohere, All models listed will work.
Vogels doesn't promise that Distill won't make mistakes.
“Remember, AI is not perfect,” he writes. “Some of the summaries we come back to contain errors that we need to manually adjust. But that’s okay because it speeds up the process. It's just a reminder that you need to be empowered and involved in the process. Critical thinking is just as important now as it was then.”
I would argue that having to “involve” summarization undermines the meaning of the automatic summarization feature. You might want to hire a stenographer. But you'll never catch Mr. Vogels badmouthing the technology his employer sells. I would definitely say that's why he's still his CTO.