Google Unveils AI-Driven ‘Ask Photos’ Feature for Enhanced Image Search
Google is testing its new Ask Photos feature, powered by Gemini AI, which allows users to search their photo libraries using natural language. This innovation, previewed at Google I/O 2024, enables users to ask specific questions about their images without needing to tag or organize them beforehand. Initial demonstrations showcased its capability to retrieve specific photos, such as license plates. While currently in testing with select users, Ask Photos aims to significantly enhance image organization and retrieval, showcasing Google’s commitment to integrating AI across its services.
Apple Unveils High-Performance AI Models Surpassing Competitors
Apple has launched a new set of open-source AI models through its DataComp for Language Models project, showcasing significant advancements in natural language processing. The models, featuring 7 billion and 1.4 billion parameters, have demonstrated superior performance against competitors like Mistral and Hugging Face. By utilizing a standardized framework for training, the larger model achieved a 63.7% accuracy on the MMLU benchmark, marking a notable improvement over previous state-of-the-art models. This initiative emphasizes the importance of effective dataset design and aims to enhance AI training strategies, although the models are not intended for direct use on Apple devices.
Mistral AI and NVIDIA Launch Multilingual 12B NeMo Model
Mistral AI, in collaboration with NVIDIA, has unveiled the NeMo model, featuring 12 billion parameters and an extensive context window of 128,000 tokens. This model excels in reasoning, world knowledge, and coding accuracy, designed to replace the existing Mistral 7B seamlessly. With an open-source approach, both pre-trained and instruction-tuned checkpoints are available under the Apache 2.0 license, fostering research and adoption. NeMo introduces a new tokeniser, Tekken, which enhances compression efficiency across multiple languages. This release aims to democratize access to advanced AI capabilities, making it a valuable resource for diverse applications.
Google Enhances Paris Olympics Coverage with AI Innovations
Google has been named the official search AI partner for Team USA, marking a historic collaboration with the U.S. Olympic and Paralympic Committee. The company will integrate AI into the U.S. broadcast of the Paris Olympics, enabling commentators to use AI for real-time explanations of competitions. NBCUniversal plans to modernize its coverage to engage younger audiences, featuring personalized AI-generated recaps narrated by AI versions of prominent commentators. Comedian Leslie Jones will also leverage Google’s Gemini AI to enhance her commentary, enriching the viewer experience as the Games approach.
OpenAI Explores Development of New AI Chips with Broadcom
OpenAI is in discussions with Broadcom and other chip designers to develop its own artificial intelligence chips, aiming to address the ongoing shortage of costly graphics processing units essential for its AI models like ChatGPT and DALL-E3. The company is hiring former Google engineers who previously worked on AI chips and plans to create an AI server chip. OpenAI is also engaging with industry and government stakeholders to enhance access to necessary infrastructure, while CEO Sam Altman seeks billions in funding to establish semiconductor manufacturing partnerships with major chipmakers.
