Gemini Live could use some more rehearsals | TechCrunch

🌈 Abstract

The article discusses the experience of using Gemini Live, Google's new AI-powered voice assistant, and the challenges it faces in providing a reliable and engaging conversational experience.

🙋 Q&A

[01] Gemini Live's Capabilities and Limitations

1. What are the key features and limitations of Gemini Live?

• Gemini Live is a more free-flowing and natural-feeling voice assistant compared to Google's previous attempts, but it still suffers from issues like hallucinations, inconsistencies, and a lack of expressiveness.
• It has a dispassionate tone and lacks the ability to adjust voice characteristics like pitch, timbre, or pace, putting it at a disadvantage compared to Advanced Voice Mode.
• Gemini Live does not have the same integrations as the text-based Gemini chatbot, limiting its functionality.
• The article highlights technical issues with Gemini Live, such as voice cutting out, difficulty recognizing responses, and the need for non-intuitive steps to get it working.

2. How does Gemini Live's performance compare to human-like conversations?

• Gemini Live maintains a polite but apathetic tone, giving the impression of handling multiple conversations simultaneously rather than focusing on the user's needs.
• The bot's tendency to confidently make up information and "gaslight" the user makes it difficult to trust its responses.
• Gemini Live's answers are often generic and not particularly useful, even when the information is factually correct.

3. What are some examples of Gemini Live's inconsistencies and hallucinations?

• The bot provided inaccurate recommendations for budget-friendly activities in New York City, suggesting a closed nightclub and a non-existent rooftop bar.
• When the author tried to stump the bot, Gemini Live contradicted its own previous statements, demonstrating its tendency to make up information.
• The bot refused to comment on political figures and elections, suggesting limitations in its knowledge and capabilities.

[02] Potential Improvements and Use Cases

1. What potential improvements or use cases are mentioned for Gemini Live?

• The article suggests that Gemini Live's utility may improve once it can interpret images and real-time video, which Google plans to add in a future update.
• The author speculates that Gemini Live could be useful for job interview preparation, but notes that the bot's feedback was generic and not particularly helpful.
• The article suggests that the text-based Gemini chatbot may currently be more useful than the Gemini Live voice assistant.