Gemini 1.5 Pro announced support for 2 million tokens (waitlist recruiting). The official blog mentions "a series of quality improvements across key use cases including translation, coding, and inference" but does not disclose any benchmarks.
The fourth model, Gemini Flash , has been added to the existing three models. This model is described as “optimized for fast and frequently required AI tasks,” and it is emphasized that it provides a capacity of 1 million tokens at a slightly lower price than GPT3.5, but the exact figures for speed have not been announced. The Gemini product line that has been released so far are as follows:
Gemini Live : “The ability to have deep, two-way conversations using voice,” leading directly into Project Astra, a real-time, video-understanding personal assistant chatbot with a two-minute demo.
Gemma 2, previously 7B and 2B, has now grown up to 27B and is a model in training that provides performance close to Llama-3-70B at half the size (fits into 1 TPU). This will also be made available for free to run locally.
Imagen 3: Google's image generation model, which reduces the burden on users by improving the understanding and interpretation of prompts compared to previous models. (It is the next generation model of the existing Imegen.)
A new hardware called TPUv6, called Trillium, has been unveiled. It is significantly better in performance than existing TPUs. (4.7x performance improvement)
And we announced the integration of AI technology across Google products, including Workspace, Email, Docs, Sheets, Photos, Search Overviews, Search with Multi-step Reasoning, Android Circle to Search, and Lens.
CNET has a 12-minute summary, so if you're interested, please refer to the video below or the summary in Release AI.
Gemini 1.5 Pro, announced at Google I/O, stands out for its improved processing speed and MMLU numbers, which are expected to significantly improve the user experience by extending the context length compared to the existing model. In addition, Gemini 1.5 Flash is impressive for its significant improvement in text generation speed while maintaining the 1M token processing capability despite being a lightweight model. This highlights the integration that utilizes Google’s powerful infrastructure and enables extremely fast and efficient response generation.
Innovative features such as Project Astra’s ability to process real-time audio/video data and generate responses are notable advances that enable real-time conversations even on prototypes like Google Glass. In addition, the rapid evolution of open-source models like Gemma is making AI research and development more accessible through collaboration with the developer community. This strategy is an example of Google’s ongoing commitment to providing better services to both users and developers.
The introduction of Context Caching has the potential to reduce repetitive inputs for long contexts, reduce costs, and greatly improve user convenience. The improvement of the interface to accommodate various inputs will contribute to diversifying and enriching the user experience. These technological advancements and innovative approaches clearly demonstrate that the technologies introduced at Google I/O are having a profound impact on user experience and the developer ecosystem.
However, I couldn't help but feel regretful. It was a huge and long presentation, but wasn't there a point that I could say "Wow"? There was no innovation as a new product, but I got the impression that there was innovation in business. It showed good aspects in terms of cost, operation, and utilization, but it may be because OpenAI or Apple took the lead in such highlights.
Subscribe to 'haebom'
📚 Welcome to Haebom's archives. --- I post articles related to IT 💻, economy 💰, and humanities 🎭. If you are curious about my thoughts, perspectives or interests, please subscribe. Would you like to be notified when new articles are posted? 🔔 Yes, that means subscribe. haebom@kakao.com
