The Weekly AI Roundup: Gemini 2.5 Pro Released, Deep Fakes In Politics, Everything is GHIBLI!

As anticipated, DeepSeek has released its new V3 model in March.It’s a huge 700 gigabyte model and of course it’s open source. It’s already available on platforms like Hugging Face and Open Router and users are busy putting it through its paces to see just how much of a step up it represents. It’s inevitable AI will increasingly play a major role in the political theatre. Social media posts have been circulating of an audio clip that is said to be US Vice President JD Vance criticising Elon Musk. However, many people that have analysed the clip have claimed that it is fake and has been generated by AI. The Model Context Protocol or MCP has been getting everyone very excited because of its adoption by many services and its potential to become a new standard for AI tool integration. Zapier got in on the action by announcing Zapier MCP that they claim will allow you to integrate more than 8,000 different services into your favourite AI assistant like Cursor or Windsurf. Meta released a new paper for a model that can create complete 3D scenes from a single image. These images are then ready to be used in virtual reality. Google has released Gemini 2.5 Pro experimental, a thinking model that Google of course claims is their most intelligent model ever. Featuring a 1 million token context window, Gemini 2.5 Pro is already available in Google AI Studio and the Gemini app. Incredibly, it scored 18.8% in the humanities last exam benchmark. The battle for the world’s greatest AI image generation really heated up today. Reve, launched by a startup of the same name, came from out of nowhere, this model immediately jumped to the top of the pile with many proclaiming it as the new king of image generation. But then OpenAI launched 4o image generation integrated into ChatGPT and Sora. This model is not only producing some amazing quality images, it’s also leveraging OpenAI’s LLMs to help understand the exact nature of the user’s request. Figure Robotics has revealed that they have made a significant breakthrough with their figure 02 robot, which now walks with a gait that they say closely resembles a human. This was achieved through reinforcement learning in a high fidelity physics simulator where extensive data was processed rapidly and then transferred directly to real hardware without additional tuning. Grok is getting integrated into the messaging app Telegram. Elon Musk has announced that the XAI tool will now be available to use in Telegram for premium users at no additional cost. Microsoft is enhancing its CoPilot suite with two new AI powered agents, ‘Researcher’ and ‘Analyst’, that use OpenAI’s advanced reasoning models. Google’s AI suite Gemini is adding access to many new Google apps and tools to make it more capable of carrying out multi step tasks across different services. The new additions include calendar, notes, tasks and photos. With the launch of OpenAI’s 4o image generation yesterday, everyone has gone image crazy. And there’s one style in particular that has been flooding everyone’s timeline the Ghibli Studio look. Even Sam Altman himself had to take a moment to reflect on how this seems to be his finest achievement. MCP, or Model Context Protocol, is a set of standards for connecting services to LLMs that is being adopted very quickly throughout the industry. With OpenAI now expressing support for the protocol, it’s likely to be the tipping point needed to make MCP the standard across AI services. Anthropic has unveiled a new research tool that they describe as a microscope for looking inside an LLM. This will help researchers and engineers to better understand the kind of thinking that an LLM like Claude is doing and also fine tune the process for better results. Pika has released a new ‘Flashback’ feature that allows you to meet your younger self. Users just upload a selfie style video and then a photo of themselves from when they were younger and Pika does the rest. OpenAI rolled out some extensive updates to ChatGPT 4o today and as a result it leapfrogged right above ChatGPT 4.5. OpenAI claimed that this made ChatGPT better at following detailed instructions, improved its capability to tackle complex coding problems, and improved its intuition and creativity. The popular AI benchmarking tool LM Arena promoted GPT4o from fifth to second place and knocked GPT 4.5 down to fourth. The Ghibli craze has not slowed down, on X, Sam Altman posted, “it’s super fun seeing people love images in ChatGPT, but our GPUs are melting.” He went on to say that they would have to temporarily rate limit some requests for users just to keep everything online. ElevenLabs has introduced a new Actor mode where users are able to record their own voice and have the AI voice replicate the exact pace, intonation and emphasis from the recording.