After wrapping up another interesting Display Week last week in San Jose, I decided to go back and read the April 2025 issue of JSID. There are quite a lot of disparate topics covered like folding mechanics, quantum dot microdisplays and VR user experience, so it was time to try something new for a slightly quicker review – Google’s Notebook LM.

It’s been around for quite a while now by the tech world’s standards (aka months…) but I’ve not yet used it for summarizing primary research. This first attempt was… eye opening to say the least. Listen to the summary below if you have ~35 minutes to spare driving to the office:

Barring the obvious issue that not all research papers were summarized and there are a few glitches here and there, this is an incredibly impressive advancement in deep tech AI summarization. Here’s a few technical issues I’ve seen:

• Feng et al. (Foldable OLED panel optimization) – pretty good! It calls out an incorrect number of optimization variables for the panel structure, but this doesn’t change materially the conclusions drawn.
  
• Guo et al. (Comfort interactions in VR) – seems to struggle referencing virtual image distance (VID) correctly throughout the entire section, but no issue correctly pulling out the right study results and implications.
  
• Song et al. (QDCC by liftoff) – this one struggled with the naming of inorganic and organic materials, as well as a misunderstanding of industry-specific acronyms usage out of context. Again, overall no impact on key takeaways especially technical results like PPI implications and CC efficiency.
  
• Zhao et al. (High bandwidth electo-optical modulator) – I’ll be honest that this one is furthest from my expertise, but there is some clear confusion of the material used for EOM construction and the actual EOM structure itself. That said, no clear issue with misrepresenting the results.

Overall minor issues don’t detract from a good summary for industry experts, but don’t trust these 100% as a layperson just yet… That said, we are seeing the worst these AIs have to offer – how much longer until we start synthesizing all this new knowledge almost exclusively through these types of summaries?