Using Google AI to Summarize My Own Paper - A Trial of Illuminate

(Translated from the Chinese version with the help of Claude.)

(It’s been a long time since I’ve dealt with academic materials. This article attempts to revisit the academic paper format.)

Abstract

Google’s latest experimental AI product can generate discussion audio from academic papers. The author used their own paper as input to try the product and found that the generated audio provides a fairly accurate summary of the paper. While the product currently has some limitations, it shows promise for the future.

Background

At a Google internal conference, there was an AI project demo called Illuminate - turn academic papers into AI-generated audio discussions. You give it the URL of one (or several) papers, and it generates a 3-5 minute audio segment in the format of a conversation between a student and teacher, discussing the main content and significance of the research. My first impression was that this seems impressive, but probably not very useful in practice? The presenter said the main uses are: 1. Some people prefer learning through listening rather than reading text; 2. This method can help illiterate populations. I thought to myself, wouldn’t illiterate people be uninterested in academic papers? Maybe they’re starting with papers because academic papers have relatively fixed formats and structures, making them suitable as machine learning materials, and will expand to other text materials later? But I thought, whatever, why not try uploading some of my previous papers to play around? The examples provided on their website are all from computer science and machine learning fields - I wonder how it would perform on chemistry papers?

Experimental Procedure

As soon as I thought of it, I did it. First, I uploaded the paper to my own website, then entered the URL into the system. After waiting a few minutes, an audio file was generated. Using blackhole to fiddle with QuickTime screen recording, I obtained the audio file (due to my slow reflexes, there’s some silence at the beginning of the audio 😂). The audio transcript was generated by https://go-transcribe.com/.

Results and Discussion

We used a paper I published in 2014 (ten years ago!) as an example. I don’t recommend everyone read the original (unless you have a special interest in palladium-catalyzed ethylene polymerization). The abstract is also a long paragraph full of numbers, letters, symbols, and jargon - quite intimidating. The graphical abstract is relatively clearer, but the information is still quite complex with a high barrier to entry.

The audio output from the tool is as follows (image generated by ChatGPT):

The entire audio is less than two and a half minutes, discussing the research subject of this paper (a class of palladium catalysts for ethylene polymerization and copolymerization), why the research is important (the resulting polyethylene is widely used in daily life), some basic related concepts (ethylene polymerization, kinetic processes), key findings (different activation energies for two kinetic processes), and main characterization methods (NMR spectroscopy). As the paper’s author, I think this dialogue serves as a qualified brief introduction for someone who might be (or pretending to be) somewhat interested in this research. Of course, the audio has some shortcomings, such as odd pronunciation of some technical terms probably due to their rarity (e.g. chelate), and one person responding too quickly after the other finishes speaking, making it sound unnatural.

Conclusion

This AI tool has a relatively high level of completion. The generated audio observed in the experiment provides quite accurate summaries of the input paper’s research theme and main findings. If more features are added in the future (such as allowing users to ask questions), it could be even more useful.

Supplementary Materials

Audio transcript:

Student: Let’s unpack a paper titled differentiation between chelate ring inversion and aryl rotation in a cf3 substituted phosphine sulfonate Palladium methyl complex. What’s the core idea here?

Teacher: Right on. This paper is a deep dive into the world of palladium complexes, specifically those with ortho phosphine sulfonate ligands. These complexes are catalysts for olefin polymerization and copolymerization.

Student: Olefin polymerization you say, that sounds pretty important.

Teacher: Absolutely. It’s a key process in making plastics and other materials we use every day. These palladium complexes are catalysts, meaning they help speed up and control the reactions that make these polymers.

Student: So what’s the big deal about this particular paper?

Teacher: Well, the authors focused on the dynamic properties of these complexes, like how they change shape in solution. They used NMR spectroscopy techniques to track these changes.

Student: NMR spectroscopy huh, that sounds pretty technical.

Teacher: It is, but it’s a powerful tool for studying molecules. In this case, they were particularly interested in two processes chelate ring inversion and aryl rotation. These are like two different ways the molecule can change shape and understanding them is crucial for designing better catalysts.

Student: I see. And what do they find out about these processes?

Teacher: They found that the barriers to these processes are different, meaning one is easier for the molecule to do than the other. This is important because it gives us clues about how these catalysts work and how we can tweak them for better performance.

Student: So it’s all about understanding how these molecules change shape, right?

Teacher: Exactly, and this paper provides some valuable insights into that.

Student: That’s exciting. It sounds like this paper could have a real impact on the field.

Teacher: I think so too, it’s a solid piece of work that adds to our understanding of these important catalysts.

Student: Definitely. It’s always cool to see new tools being used to tackle these problems.

Teacher: I agree.

Student: And this closes our discussion of differentiation between chelate ring inversion and aryl rotation in a cf3 substituted phosphine sulfonate Palladium methyl complex. Thank you.