This week, the world of artificial intelligence (AI) was in the spotlight thanks to several loud events from Openai, including Sam Altman, starting the Advanced Voice Mode, rumors about the new 5GW processing center, as well as personnel permissions and reorganization in Company. However, in addition, other technological giants continued to develop their AI projects, representing new models and research. Below are key news in AI in the last days.
Google gemini updates
google announced new versions of its Gemini model, releasing updated models Gemini-1.5-Pro-002 and Gemini-1.5-Flash-002. These versions received improvements in performance, including success in processing mathematical tasks, long contexts and visual tasks. Google announced an increase by 7% performance on the MMLU-PRO test and an improvement by 20% in mathematical tasks. An important addition to the updates was a significant decrease in the cost of using the model: the cost of input tokens decreased by 64%, and the weekend – by 52% for requests to 128,000 tokens. 2.000 queries per minute, and for Gemini 1.5 Pro – 1.000 queries. Steel models are twice as soon as possible and show a 300% lower delay compared to previous versions. This makes the use of these models more affordable and effective for developers.
launch of Meta* llama 3.2
meta also I was pleased with the news , releasing the update of my Llama model to version 3.2. Among the new products are models with support for image processing and visual data, with dimensions of 11 and 90 billion parameters, as well as text models of smaller volume (1 and 3 billion parameters) designed for mobile and peripheral devices. META claims that new models for working with images successfully compete with leading closed models in the problems of recognition and understanding of visual information. Empty text models show high results in comparison with similar solutions.
of one of the main novelties was the release of official Llama Stack distributions, which simplify the process of developing and expanding models on various platforms. Llama 3.2 models, as before, are available for free download with a license that limits commercial use.
Google Deepmind and Alphachip to accelerate the development of chips
Google DeepMind I announced