Fresh from its breakout success with NotebookLM, the mighty Google machine has again swung into gear with another offering from its research laboratories.
Google Illuminate is a similar podcast maker product to NotebookLM in many ways, but completely different in one crucial feature — its focus is purely on the podcast.
Both of these AI tools are designed to make it quicker and easier to distill knowledge into friendly digestible chunks and present it in a way that is convenient for humans.
Illuminate is currently only available to a select group with a waitlist to gain wider access. It isn’t clear if it will ever become more than just an experiment.
What is Google Illuminate?
Google Illuminate is a simple podcast generator that uses AI to create a conversation, similar to the podcasting feature in NotebookLM. The principle behind these new tools is the fact that AI is spectacularly good at reading through large volumes of text, video or audio and then producing insightful summaries.
What gives Google the edge in this field is that its Gemini AI model has a huge context window of double its rivals, at 2 million tokens. A context window is essentially how much information the AI can hold in a single session before it starts to forget or degrade.
This means that where other models, from OpenAI or Anthropic, struggle to handle large amounts of data in one go, Gemini can ingest without a problem. Perfect for deep-dive research into knotty topics, or arcane scientific subject matter.
Illuminate is the second product — or experiment as Google calls it — to take advantage of this technology lead, and it does so with style.
How do you use Google Illuminate?
To use the Google Illuminate service you sign up and log in (assuming you’ve got through the waitlist), and then hit the Start Generating button.
At this point, you are presented with the option to search the AI for a topic, upload one or more PDF documents on a subject, or drop in a relevant web address. Or a combination of the three. Just feed the machine and move on to the next step.
Once you’ve finished feeding the machine, press the Generate button and wait while a cleverly assembled audio podcast is created as a chat between two people on the subject of your choice.
The result is almost identical to NotebookLM, although it does use different voices. And there’s one major interface difference. The user is given the chance to tweak the output settings to deliver a range of different results.
So, for example, you can choose to produce a podcast suitable for beginners rather than experts, in a casual tone and for a short length of time. Or vice versa. It’s very clever and a nice way to again make knowledge more accessible.
There is a major limitation to Illuminate
There is a catch to Illuminate. Unlike NotebookLM, Illuminate is strictly restricted in the material it will process. Specifically, scientific research from the arxiv.org website, which is a free and open-access science portal run by Cornell University in the US.
The site contains around 2.4 million scholarly articles in the sciences, including physics, math, biology, economics and all the usual suspects. It’s not peer-reviewed, but is well respected for releasing timely and important scientific research on a freely distributed basis.
This rigid scientific focus is a deliberate feature of Illuminate, and is aimed at differentiating it from NotebookLM in terms of delivery.
It’s not hard to see a future where this type of technology will be used, not just for one specific research site, but for all research resources across the world. This could easily result in an extremely rich resource for academics, students, researchers and the general public. A global AI powered encyclopedia unlike anything that’s ever been produced before.
In a recent lengthy interview on YouTube, Raiza Martin, lead manager on the NotebookLM product, outlined the history of the product (which started as a Google 20% project in Google Research Labs), and how it might in the future morph into a powerful knowledge factory. This would mean accepting data in just about any format, processing it at a massive AI scale, and outputting the results in a wide variety of formats.
You could, for example, take dense research, query it and output, say, a professional-style video documentary with the most important points highlighted for further action.
For now, Illuminate, and its cousin NotebookLM, are clearly works in progress. It’s all very compelling, but there are also gaps where further development is needed. But no matter how things go, this type of product seems to have a promising place in the world’s future AI toolbox.