Google has widened availability of its Gemini Live function to all users on Android free tier, no longer limiting it to those with an Advanced subscription. Originally introduced with the Pixel 9, Gemini Live enables dynamic voice interactions where users can interject during voice responses.
Features and Usability of Gemini Live
Gemini Live can be activated via a circular waveform icon with a sparkle found on the Gemini overlay or by launching the main application, providing an immersive fullscreen experience complete with “Hold” and “End” buttons. The tool runs unobtrusively in the background, allowing multitasking or screen locking without disruption. Conversations conducted are saved as text transcripts in the app's history section for easy retrieval or re-engagement.
We're starting to roll out Gemini Live in English to more people using the Android app, free of charge. Go Live to talk things out with Gemini, explore a new topic, or brainstorm ideas. Keep an eye out for Gemini Live in the Gemini app 👀 pic.twitter.com/0VL0c7E6Gw
— Google Gemini App (@GeminiApp) September 12, 2024
The feature enhances user interaction with a selection of ten unique vocal choices accessible under “Gemini's voice” in settings. Voices, offering a variety of tonal styles and accents, include:
- Nova: Moderately calm
- Ursa: Involved, with a mid-range tone
- Vega: Lively and higher toned
- Pegasus: Deep, active
- Orbit: Vibrant, deeper
- Lyra: High and bright
- Orion: Deep and luminous
- Dipper: Deep, actively engaged
- Eclipse: Mid-range energy
- Capella: High tone with a British accent
Limitations and Future Developments
Currently missing is support for Gemini Extensions like Gmail and YouTube Music, although such integrations are expected later. Similar to OpenAI ChatGPT's voice feature, it allows spoken questions and interruptions, promoting smoother exchanges.
Gemini Live is being rolled out for English-speaking Android users, with future updates to incorporate additional languages and iOS capability expected. Previously exclusive to Advanced subscribers, this move indicates Google's intention to make the feature more broadly available.
OpenAI o1 Brings Deeper Reasoning to AI Models
Google's AI rival, OpenAI, has taken its models a step further this week with the release of it o1 model, which is known internally as Strawberry. The o1 model represents a significant departure from OpenAI's traditional development methodologies. Instead of solely relying on pattern recognition from extensive datasets, the model employs a novel optimization algorithm and a distinct training set.
By integrating reinforcement learning, it addresses queries through a process known as “chain of thought,” thereby augmenting its reasoning capabilities. OpenAI acknowledges that while accuracy has improved, the challenge of AI generating inaccurate information, commonly referred to as hallucinations, remains unresolved.
Regarding accessibility and pricing, both ChatGPT Plus and Team subscribers can now utilize the o1-preview and o1-mini models. OpenAI plans to extend access to Enterprise, Edu, and eventually free ChatGPT users, although the specific timeline remains indeterminate. Developers will incur higher costs for accessing the o1 version, with pricing set at $15 per million input tokens and $60 per million output tokens. This represents a substantial increase compared to GPT-4o's fees of $5 and $15.
