From Google's new large language model to H&M using AI models: A few big AI innovations take centrestage

Artificial Intelligence

Things move fast in the world of technology. Ever since AI became the buzzword, things have been moving even faster. But even by those ‘rapid’ standards, this week awfully busy with big players showcasing advancements across multiple domains. From new models to transformative applications, the AI landscape saw a lot of activity. Here are some of the big announcements made by Google, DeepSeek, OpenAI and others

On March 26, OpenAI launched GPT-4o’s native image generation feature, enabling users to create detailed visuals directly within ChatGPT. The feature has sparked creativity online, with users transforming images using the Ghibli effect. From Elon Musk and Donald Trump’s ‘bromance’ to historical images getting a makeover, social media was abuzz with ChatGPT’s new feature. Furthermore, the image generation feature does a fine job of rendering text within images, refining designs through conversation, and handling complex prompts involving multiple objects.

Chinese AI startup DeepSeek unveiled its latest model, Deepseek-V3-0324. This open-source large language model boasts 685 billion parameters and employs a Mixture of Experts (MoE) architecture, activating only the most relevant parameters for efficiency. The model scores high in coding, mathematical problem-solving, and reasoning, outperforming competitors in benchmarks like MATH-500 and AIME 2024. Its release on Hugging Face has made it accessible to developers worldwide.

Google introduced Gemini 2.5 Pro, its most advanced AI model yet. This "thinking model" integrates reasoning capabilities directly into its base architecture, eliminating the need for specialized modules. Gemini 2.5 Pro leads benchmarks such as Humanity’s Last Exam and GPQA Diamond, showcasing its prowess in coding, mathematics, and multimodal tasks. With a million-token context window and multimodal support, the model is accessible via Google AI Studio and Gemini apps.

Fashion giant H&M announced its use of AI-generated digital twins for advertising campaigns. These lifelike avatars, created from real models, streamline content creation and reduce reliance on traditional photoshoots. H&M made its intention clear to utilise artificial intelligence-generated doppelgängers in certain social media posts and marketing campaigns, contingent upon the consent of the models involved.

Researchers unveiled ECgMLP, an AI model capable of detecting endometrial cancer with 99% accuracy. This breakthrough surpasses previous systems and extends to other cancers like colorectal and breast. The model’s ability to process visual data and highlight risks, promises earlier interventions and improved outcomes.

Alibaba released Qwen2.5-Omni-7B, a multimodal AI model designed for edge devices like smartphones. With capabilities spanning text, images, audio, and video, the model supports applications such as real-time voice assistance and customer service. The model is open-source and will be accessible via Github and Hugging Face.