Google continues to actively invest in AI advancements, but its offerings haven’t always been consistent. The Pixel Screenshots app is a success story, offering intelligent analysis and organization of screenshots for easy retrieval through conversational search. However, the AI-enabled Gemini summaries for Google Recorder have fallen short of expectations.
Gemini Nano’s Finicky Nature
The on-device Gemini Nano LLM powering the feature has been quite particular. It promises real-time transcription summaries, but it often encounters errors for recordings that are either too short (under a minute) or too long (over 15 minutes). Short recordings aren’t as significant of a concern, but the limitation on longer recordings is a drawback, considering the greater utility of summaries for lengthier content.
Pixel 9 Series Brings Improvements
According to a new report by Mishaal Rahman in Android Authority, the Pixel 9 series addresses some of these limitations. These devices utilize an upgraded LLM called Gemini Nano with multimodality, enabling transcripts and summaries for much longer recordings. While the exact limit remains undisclosed, the report indicates successful transcription of a 41-minute recording—a substantial improvement over the Pixel 8 series’ 15-minute cap.
Lingering Limitations and Future Uncertainties
It’s important to acknowledge that even with the enhanced LLM, recordings surpassing an hour still trigger the “Transcript is too long” error. Additionally, there’s no current indication from Google that Gemini Nano with multimodality will be available for older Pixel devices. Thus, the ability to transcribe lengthy recordings on-device might remain exclusive to the Gemini Nano.
Google’s pursuit of AI excellence is evident, but its tools sometimes struggle to deliver on their full potential. While Gemini summaries for Google Recorder have shown progress with enhanced capabilities on Pixel 9 devices, limitations persist. The hope remains that future updates will further refine this feature, enabling seamless and comprehensive summaries for recordings of all lengths, regardless of the Pixel model.
Add Comment