Google released 22 AI tricks across 7 models
At the annual Google I/O developer conference, Google presented a compelling case for the potential of AI. During the opening ceremony, which lasted less than two hours, Google CEO Sundar Pichai and a group of Google executives mentioned ‘AI’ a total of 121 times. In a highly anticipated speech at the I/O conference, Google DeepMind co-founder and CEO Damis Hassabis placed second only to Pichai in terms of speaking time. This indicates that the company's focus on AI has intensified.
Yesterday, OpenAI launched a surprise attack on the traditional voice assistant flagship model GPT-4o, which is seen as the ‘world's first brother of AI + search brother’ of Google. This is likely to lead to a re-alignment of the AI track ‘big model king’ rankings. Google will be the first to lead the AI track. In a bold move, Google unveiled 22 AI innovations, including the launch of 2 million tokens of ultra-long context Gemini 1.5 Pro advanced version, Gemini 1.5 Flash lightweight model, universal AI Agent, high-quality text-generated graphical model Imagen 3, AI music creation tool Music AI Sandbox, 70-second video generation model Veo, and the first visualisation model. The Generation Model Veo, the first visual language open model PaliGemma, and many other models were also showcased, along with the next generation of Gemma 2 big models.
Other significant developments include the introduction of 6th generation TPUs, AI infrastructure, new AI search features, Google Workspace app Gemini features, Gemini Live multimodal features, Gemini customisation features, Gemini Advanced, Draw Circle and Search features, new Gemini Nano features, Android Gemini collaborations, AI Assist Red Team technology, extended and open source SynthID text watermarking, and more.
Google also showcased a range of AI systems, including RT-2, which translates vision and language into robotic actions, SIMA, which navigates complex virtual 3D environments, and AlphaGeometry, which solves OU problems. The launch commenced with Pichai stating that over 1.5 million developers are currently utilising the Gemini model, and that Google has two billion users with products that utilise Gemini. The Google-launched apps are available on Android and iOS to interact directly with Gemini, and over a million individuals have registered to test it in three months.
Google Search represents the pinnacle of generative AI for the vastness of human curiosity. It is the most exciting development in our search space. Pichai made a passionate statement at Google I/O. In the wake of OpenAI ChatGPT, Google, the former AI big brother, was forced to reduce its lead in the AI big model race due to a slow response and the 'car' event.
Subsequently, Google rekindled its AI development efforts, constantly refining the Gemini model, and at this year's I/O conference, showcased a range of cutting-edge AI capabilities. Google has made a series of innovations, including Gemma 2, Gemini 1.5 Flash, Imagen 3, Veo and other new models. It has also upgraded its search capabilities through text, voice, video, images and other multi-modal search. Additionally, it has developed custom AI dedicated chips and infrastructure for the next generation of AI models and agents to provide faster and lower latency training and service power. A series of achievements that demonstrate the company's technical expertise and extensive market reach.
In the competitive field of generative AI, OpenAI is not currently in a dominant position. There are still a number of variables that will determine the ultimate winner.
