⚡Google Launches Gemini 2.0 Flash

Google unveils Gemini 2.0 Flash, its new flagship AI model

Key Points 

• Gemini 2.0 Flash is faster and more versatile, with features like real-time multimodal capabilities and third-party tool integration.

• Developers can explore Gemini 2.0 Flash now via APIs and Google platforms, with a full release set for January.

🚀News - Google has unveiled its next-generation AI model, Gemini 2.0 Flash, stepping up competition with OpenAI. 

Announced on Wednesday, this upgraded model goes beyond text generation to natively create images and audio. It also integrates with third-party tools, enabling features like Google Search access, code execution, and more.  

✨What else is new? Gemini 2.0 Flash is a major leap from its predecessor, 1.5 Flash, which was text-focused and limited in performance. This new model is faster, smarter, and better equipped to handle complex tasks, including coding and image analysis. According to Google, it outperforms the Gemini 1.5 Pro in speed and accuracy, now taking its place as the flagship Gemini model.  

A standout feature is its ability to create and edit images while understanding photos, videos, and audio files. For instance, it can analyze recordings to answer specific questions like “What did he say?” The model also offers customizable audio generation, allowing users to choose from eight voices tailored for various accents and languages, with options to adjust speed or even add a playful tone like a pirate.  

To address concerns about misuse, Google is using SynthID technology to watermark all AI-generated audio and images.  

👨🏻‍💻How to get started - The full version arrives in January, but developers can start experimenting today via the Gemini API, AI Studio, and Vertex AI. While some features are reserved for early access partners, the Multimodal Live API is now available for creating apps with real-time audio and video capabilities. 

Future updates promise integrations across popular Google tools like Android Studio and Chrome DevTools.  

Harvard and Google to release new high-quality AI training dataset funded by OpenAI, Microsoft

Key Points 

• Harvard’s new dataset contains nearly 1 million public-domain books that are no longer under copyright.

• The dataset is aimed at leveling the playing field by giving smaller AI players and researchers access to content usually available only to tech giants.

👨🏻‍🏫News - Harvard University has unveiled a new, high-quality dataset containing nearly 1 million public-domain books, available for anyone to use in training large language models and other AI tools. 

This initiative, created by Harvard’s Institutional Data Initiative and backed by funding from Microsoft and OpenAI, includes books scanned as part of the Google Books project that are no longer under copyright. 

The dataset is significantly larger than the widely-known Books3 dataset, used for training models like Meta’s Llama. It covers a wide range of genres, languages, and decades, featuring both literary classics and niche texts like Czech math books and Welsh dictionaries. The exact distribution method for the dataset is still being worked out, with Google supporting the project’s public release.

📚Why this matters - Greg Leppert, executive director of the Institutional Data Initiative, explained that the project aims to "level the playing field" by providing public access to valuable, curated content typically only available to large tech companies. 

He believes that this dataset, when combined with licensed materials, could help smaller players and individual researchers build their own AI models. Leppert likens it to the role Linux has played in powering much of the world’s computing, emphasizing the importance of open access to foundational data in AI development.

📜See also - In addition to the book dataset, the Institutional Data Initiative is collaborating with the Boston Public Library to scan millions of public domain newspaper articles.