On GenAI-Based Enterprise Productivity Innovation, Google Notebook LM, and the Voice Technology Tipping Point

Recent updates in the AI field have been so overwhelming that it’s becoming increasingly difficult to grasp their significance and importance. Amidst all this, there was an important update I had overlooked: Google’s Notebook LM (https://notebooklm.google/). Until now, Google’s Gemini-related updates hadn’t left a strong impression, but the release of Notebook LM has made me reconsider Google’s capabilities.

According to a recent announcement from Microsoft (https://news.microsoft.com/annual-wti-2024/), more than 75% of employees within companies are already using LLMs. The noticeable improvement in productivity has naturally led people to use them without any push to do so. In addition to individual use of LLMs, systems centered around LLMs are also being rapidly discussed and implemented. For example, there are GenAI business solutions (containing knowledge specific to a team or organization), GenAI portals (similar to the concept of OpenAI’s GPTs), and Multi-AI agent models (where different AIs can converse with each other), among others.

One clear pattern among these numerous attempts is the implementation of AI solutions that can “summarize, translate, search, respond, and aggregate the documents, code, data, and materials that I or my team provide.” When used individually, this resembles Notebook LM, and when turned into a chatbot for multiple users, it resembles GPTs. However, the bottleneck here is how well AI can "autonomously" read and understand various types of materials. These materials are too diverse—ranging from website links and images to PDFs, Word documents, and Excel files—so the user experience varies greatly depending on how upload management and UX are handled. For example, MS365 Copilot had some limitations in this area, but Google’s Notebook LM seems to have solved this remarkably well. Its direction and product completeness are excellent. In particular, Notebook LM demonstrates the ability to handle multimodal data consistently, analyzing both text and images simultaneously to understand the context of a document and accurately extract necessary information for the LLM. The experience feels seamless.

Lastly, when I saw the recently updated feature that creates a podcast based on uploaded materials, it wasn’t the feature itself that surprised me, but rather the feeling that the interface of “voice” had crossed a tipping point. We had thought the inflection point would be when, after hanging up the phone, we couldn’t tell whether we had spoken to an AI or a human—and now, it seems we’ve reached that point. It’s time to think deeply about the changes this technology will bring.

The advancement in voice technology is one of the most striking aspects of recent AI innovations. In particular, voice-based interfaces have now surpassed a tipping point, significantly transforming the user experience. For example, natural language processing (NLP) and deep learning-based text-to-speech (TTS) technologies have evolved to a level where they are almost indistinguishable from human voices. Furthermore, automatic speech recognition (ASR) technology can now convert speech to text with high accuracy, even in noisy environments. Going forward, screen-centric interfaces are likely to undergo significant changes (or expansions), and I believe we need to explore the possibilities for evolving current products and services accordingly.