HomeArtificial IntelligenceFor some reason, Amazon's CTO created a gathering aggregation app

For some reason, Amazon's CTO created a gathering aggregation app

How does Amazon CTO Werner Vogels – a person price untold thousands and thousands who completely bought the small Airbnb in the middle of Amsterdam where he had been living throughout the COVID-19 pandemic – spend his days? What it looks like: Developing AI-powered meeting summarization apps. Imagine that.

In one post This week on his personal blog, Vogels describes Distill, an open source app he developed together with his “OCTO” (Office of the CTO) team to transcribe and summarize their conference calls. Distill takes an audio recording of a gathering (in formats like MP3, FLAC, and WAV), analyzes it, and creates a summary together with a listing of things to finish. Optionally, this summary and list will be spit out to platforms like Slack via custom integrations.

A sample summary from the Vogel's Distill meeting summary provided by Amazon Tech.
Photo credit: Distill

As you may expect from an app from Amazon's CTO, Distill conspicuously relies on paid Amazon services to do the computationally intensive work. AWS Transcribe performs Distill transcription; Amazon S3 provides storage for the meeting audio files; and Bedrock, Amazon's generative AI development suite, handles the synopsis.

But why create a gathering summary when there are countless tools that will serve this purpose? Well, I actually have to assume Vogels thought, why not? He has tons of resources and seemingly enough free time for hobby programming projects. According to the blog, he’s already attempting to port Distill's codebase from Python to Rust. (Being CTO is a pleasant job in case you can get it.)

What's unique about Distill is that you would be able to select which AI model does the meeting summarization. By default it’s Sonnet, a mid-range model from Anthropic's Claude 3 family. (Amazon's large stake in Anthropic could have something to do with this design decision.) But any model hosted in Bedrock will work, like Meta's Llama 3 and models from AI startups Mistral, AI21 Labs, and Cohere.

Vogels doesn't promise that Distill won't make mistakes.

“Remember, AI will not be perfect,” he writes. “Some of the summaries we get back… have errors that require manual correction. But that's okay since it still quickens our processes. It's only a reminder that we want to proceed to be critical and involved in the method. Critical pondering is as essential today as ever.”

I might argue that the must be involved in summarization defeats the aim of automatic summarization. You might as well hire a stenographer. But you'll never catch Vogels talking badly in regards to the technology his employer sells. And I bet that's why he's still CTO.


Please enter your comment!
Please enter your name here

Must Read