One of the primary concerns among AI skeptics is the substantial energy consumption associated with artificial intelligence technologies. Traditional data centres require immense amounts of power, not only to operate the servers but also to cool the equipment, preventing overheating. This cooling process often involves energy-intensive air conditioning systems, contributing significantly to the overall energy footprint of AI infrastructure.
To address these challenges, innovative solutions are being explored globally. While concepts like space-based data centres have been proposed, China has taken a pioneering approach by developing underwater data centres. This allows the centre to be naturally cooled by its surroundings, eliminating the need for energy-intensive cooling fans.
Google has unveiled Gemini 2.5, the most advanced AI model it has developed so far. The 2.5 release includes an experimental version of Gemini 2.5 Pro, which has outperformed all other models in several benchmarks, securing the top spot on LMArena by a significant margin.
Gemini 2.5 models are designed to think and reason, carefully analysing information before responding. This capability results in improved performance and greater accuracy. In AI, reasoning goes beyond simple classification and prediction – it involves the ability to analyse data, draw conclusions, understand context, and make well-informed decisions.
Over time, Google has focused on enhancing AI reasoning through methods like reinforcement learning and chain-of-thought prompting, which culminated in the introduction of the Gemini 2.0 Flash Thinking model. With Gemini 2.5, Google has taken a significant step forward by enhancing the core model and refining its post-training processes. These improvements allow the model to tackle more complex problems and support smarter, more context-aware agents.
In a bold (and slightly dystopian) leap forward, Italy’s Il Foglio has unveiled the world’s first AI-generated newspaper edition. The four-page supplement, Il Foglio AI, was produced without human-written articles, relying entirely on artificial intelligence for everything from headlines to analysis. While journalists were still involved in guiding the AI and reviewing content, the edition itself was a demonstration of how far AI has come in generating news. Covering topics such as U.S. politics, Russia, and Italy’s economy, the project aimed to explore the potential role of AI in journalism.
OpenAI has recently introduced GPT-4o, an advanced multimodal AI model integrated into ChatGPT, allowing users to generate realistic images directly within the chat interface. This feature allows for the creation of photorealistic images, offering a more intuitive and creative user experience. However, due to overwhelming demand, OpenAI has implemented temporary rate limits on image generation requests to manage GPU resources effectively.
It resulted in the internet being taken over by images recreating the iconic Studio Ghibli style. The iconic Japanese cartoons have been everywhere of late and it has raised questions of copyright infringement, ethical concerns, and the role of AI in creative industries, particularly as AI models are increasingly capable of mimicking the distinctive art style of legendary creators like Hayao Miyazaki.
H&M has recently announced plans to create AI ‘twins’ of 30 models for use in marketing and social media. This move allows models to appear in multiple campaigns at once, but raises significant concerns about exploitation and the potential impact on human workers in the fashion industry. While the models involved will be compensated, critics argue that AI could replace jobs in areas like photography and makeup.
Unions and industry experts worry that this trend could erode human creativity and increase exploitation of models, as digital replicas may be used without fair compensation. Despite these concerns, AI’s potential for cost savings and diversity in modelling is also highlighted, with some arguing it allows brands to showcase a broader range of looks. However, with regulations like the Fashion Workers’ Act and the upcoming EU AI Act, the fashion industry is facing a crucial moment to balance innovation with the protection of human labour and creativity.
At NVIDIA’s recent GPU Technology Conference (GTC), the company unveiled several significant advancements in AI and robotics. The Spectrum-X and Quantum-X networking chips aim to enhance AI data centres by connecting millions of GPUs across various sites, significantly reducing energy consumption.
DGX Spark and DGX Station are new personal AI supercomputers powered by NVIDIA’s Grace Blackwell platform, designed to assist developers in prototyping and running large AI models on desktops.
Additionally, NVIDIA introduced Isaac GR00T N1, the world’s first open-source humanoid robot foundation model, developed in collaboration with Google DeepMind and Disney Research. GR00T N1 incorporates a dual-system approach for both fast and slow thinking, akin to reasoning models, and utilizes Newton, an open-source physics engine, to enhance robotic capabilities.
Despite it being a slower month in terms of AI developments, as you can see, the field is developing at an impressive rate and it is important to stay in the loop. You can read the other monthly reviews here.
