Full post at https://blog.stephenturner.us/p/illuminate-preprints-with-an-ai-generated-podcast

Google has a new experimental tool called Illuminate (illuminate.google.com) that takes a link to a preprint and creates a podcast discussing the paper. When I tested this with a few preprints, the podcasts it generated are about 6-8 minutes long, featuring a male and female voice discussing the key points of the paper in a conversational style.

There are some obvious shortcomings. It doesn’t know how to pronounce words that aren’t real words (for example, bioRxiv, or Heng Li’s new Ropebwt3), and like many text-based genAI tools, it overuses the word “delve.” And, when I gave it my recent paper describing biorecap (blog post, paper), it delved into a discussion on generative AI ethics that I never wrote about in the paper. But, aside from these few quirks, I actually enjoyed listening to the audio it produced.

I used Illuminate to generate podcasts discussing a few preprints on arXiv quantitative biology that caught my attention lately, or in the case of biorecap and pracpac, those that I authored.

The full podcast at the top of this post has all six of these preprints together, timestamped with chapters if you’re listening to this in a podcast app. Alternatively, you can listen to each individual paper below.

biorecap: an R package for summarizing bioRxiv preprints with a local LLM (https://arxiv.org/abs/2408.11707) 

BWT construction and search at the terabase scale (https://arxiv.org/abs/2409.00613)

Genomic Language Models: Opportunities and Challenges (https://arxiv.org/abs/2407.11435)

Near to Mid-term Risks and Opportunities of Open-Source Generative AI (https://arxiv.org/abs/2404.17047)

Guidelines for releasing a variant effect predictor (https://arxiv.org/abs/2404.10807)

pracpac: Practical R Packaging with Docker (https://arxiv.org/abs/2303.07876)

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit blog.stephenturner.us