https://illuminate.withgoogle.com/home
Illuminate is an experimental technology that uses AI to adapt content to your learning preferences. Discussions are generated with AI voices and are grounded in published academic papers.
I listened to the conversation about this paper on the topic of large language models that encode clinical knowledge. Conversations are a few minutes long and in my opinion, did an impressive job at taking an academic paper and turning it into something that sounds like a short podcast.
There are a few settings I’d like to be able to tweak if this eventually becomes a product:
- Length of the conversation. You may find three minutes isn’t long enough for a paper that’s more relevant to you, so dialing the duration up or down, depending on context, would be a nice feature.
- Tone of the models. The one I listened to came across as ‘morning radio’-type conversation, which was fine. But it might be nice to tell the model if you want something less casual, or more humorous, and so on.
- Factfulness (I’m not sure what else to call this). I get the impression that the models do a bit of context setting, which is really valuable for some topics. But in general, I’d prefer to just get the facts. Being able to tell the model if you want it to be more accurate or more creative (a bit like you can do with Copilot right now) would be cool.
- Playlists. Imagine being able to feed the model a playlist of papers (you could do this with a browser-extension; right-click on a link to a paper and ‘share to Illuminate’) that you could listen to as a podcast to catch up on the latest papers in your area.
I’d love for this to be developed into a full product.
Illuminate is currently in private beta, so you need to join the waitlist to experiment.