On May 21, Beijing time, the annual Google (GOOG) Developer Conference (Google I/O) was held in the United States. Google demonstrated its latest progress in the field of AI, from upgrading search engines to launching generative content tools, as well as hardware updates.
Google Gemini2.5 full series of heavy updates
This time, Google brought an upgraded version of the Gemini 2.5 model, which significantly improved multimodal understanding and reasoning capabilities. Among them, the Deep Think mode introduces an enhanced reasoning mechanism that can consider multiple possibilities before answering, improving performance in mathematics, programming and multimodal tasks.
In terms of search engines, Google introduced the “AI mode” to provide users with a more conversational and contextually understood search experience. This mode uses the Gemini model to handle multiple queries and provide deeper and more diverse search results. In addition, Google is also testing “deep search” and real-time visual data features, and plans to integrate them into regular searches.
In addition, following the release of Imagen3 in December last year, Google’s text-generated AI model ushered in the latest version Imagen4 at this I/O conference, and the video generation model was updated to Veo3. The new generation of Imagen models is 10 times faster than the previous generation and provides more realistic visual effects and better text and typography effects.
As Google DeepMind CEO Demis Hassabis said, Google will continue to double its investment in the breadth and depth of basic research, and is working hard to expand our strongest multimodal basic model Gemini 2.5 Pro, committed to inventing the next major breakthrough necessary to achieve general artificial intelligence (AGI).
Apple has made a big move to open up large model permissions to developers
At the same time, well-known technology reporter Mark Gurman wrote a new article saying that Apple (AAPL) is preparing to allow third-party developers to use its artificial intelligence models to write software. The new move will allow developers to integrate Apple Intelligenc’s technology into specific functions and enhance the attractiveness of its devices.
According to reports, Apple is developing a software development kit (SDK) and related frameworks so that external developers can build AI features based on Apple’s large language model. This plan is expected to be announced at the Worldwide Developers Conference (WWDC) on June 9.
Gurman analyzed that this move is part of Apple’s broader strategy to catch up with its competitors in the field of generative AI. Although Apple launched Apple Intelligence last year, its initial functions have not been widely used, and other AI platforms are still more powerful.
Today, from the competition of large models, major manufacturers have launched model platforms and accelerated the competition for ecological level entrances. When DeepSeek appeared, due to its leading technology, it easily acquired a large number of users without spending any marketing expenses, stimulating the competition for entrances in the vertical ecological field.
WiMi actively participates in multimodal AI competition
According to the data, Wimi Hologram Cloud Inc (WIMI), as a company focusing on holographic technology and artificial intelligence, has clearly increased its layout in the field of multimodal AI models in recent years. Focusing on multimodal AI as an important development direction of artificial intelligence, it accelerates the integration of multiple data forms such as text, images, voice, and video, improves the understanding and interaction capabilities of complex scenes, and deepens innovation breakthroughs in vertical fields.
According to public information and technical dynamics, WiMi develops AI models that can process holographic images, voice commands, and environmental sensor data in real time through multimodal data fusion to improve the interactivity of virtual content. In addition, it develops AI visual generation technology, uses generative AI to automatically generate high-precision images, and combines AI algorithms to improve real-time rendering efficiency and reduce the cost of AI model content production.
End
In fact, this year’s I/O conference has almost achieved a full update of all of Google’s current AI. It has only been one year since the last developer conference, which makes people sigh at Google’s passion for leading the current AI craze. In this regard, professionals commented that in the pursuit of AGI, every step taken by Google has attracted the attention of the world. The progress of Gemini will help Google narrow the gap with OpenAI and mark the company’s entry into a new stage of innovation.
