The Weekly AI Roundup – April 21-27
A new startup called Cluely has just released a promotion for its new AI service that might have both of them beat. The video shows a young man who is trying to cheat on a date by claiming to be 30 years old, where Cluely is guiding him through the deceptive process.
Sand AI has released a new open source video generation model that called Magi-1. It boasts a staggering 24 billion parameters, making it capable of full length cinematic videos. It does this with its capacity to create infinite video extensions along with precise control over timing and high quality output without the need for clip stitching.
The world’s first robot marathon was held in Beijing. More than 20 humanoid robots took part in the 21 kilometer race and with a time of 2 hours and 40 minutes, the Tiangong Ultra robot was declared the winner.
Testing Catalog has reported that Google are working on a new visual interface for AI Studio that allows developers to see the output of the code they are developing live as it is being constructed.
In a video put out today by Microsoft’s CEO Satya Nadella, he explains how AI is accelerating advances in healthcare and in particular disease diagnostics. The video demonstrates their dedicated medical diagnostic tool dxgpt, an AI tool designed to diagnose rare diseases where a potential diagnosis can be achieved in minutes rather than months or years.
An incredibly impressive startup story has emerged with a text to speech model called dia. Nari Labs was started by two undergrads who loved the podcast feature in Notebook LLM so much that they decided to build their own text to speech model.
Even football stars aren’t safe from the rise of the robots. The booster T1 humanoid robot from Booster Robotics is seen showing off its soccer skills with a variety of tricks and shots that are quite impressive.
OpenAI has announced that image generation will now be available via its API. They’ve launched GPT Image One that will allow developers to access the power of ChatGPT’s image generation capabilities.
Perplexity has launched its iOS voice assistant that enables users to take on multi app tasks like booking reservations, sending emails and playing media. This builds on the initial voice mode that was launched earlier this year.
Grok has also been making some major Updates to their iOS app. The latest version now has a Vision mode similar to Gemini Live, Multilingual Audio and Real Time Search in Voice mode.
Hailuo has added a new character reference feature to its video generation model. This allows users to transform a single image into dynamic characters with adjustable angles, poses and cinematic lighting.
President Trump has signed an executive order to integrate AI education into K12 schools, aiming to prepare students for an AI driven future. The Trump administration has been keen to push the US as the major player in artificial intelligence, and Donald Trump said, “we have trillions of dollars being invested in AI”.
Lovable is a popular AI coding platform and today they have announced a major upgrade to Lovable 2.0. The update introduces collaboration with other users, enhanced security checks, a smarter chat agent, a more comprehensive dev mode, and the ability to make UI edits directly on the canvas.
Perplexity announced that its app will be pre installed on all new Motorola phones and will be the default AI assistant for users.
The video generation model Higgsfield has announced a new, cheaper, faster Turbo model and along with it, seven new cinematic camera motions. These preset camera motions are at the heart of Higgs Field’s success, and the latest additions take their Library up to 80. The motion that has generated some of the most viral videos so far is the new face punch effect.
OpenAI’s announced a new lightweight version of their Deep Research tool, powered by the o4-mini model. They’re expanding access to free users and paid subscribers, with free users getting five tasks per month and paid tiers receiving higher limits and less restrictive results.
Gemini 2.5 Pro has been playing Pokemon Blue and has managed to reach the 8th badge in 600 hours.
This showcases its ability to handle complex sequential decision making in a gaming environment. You can even watch Gemini’s progress on a Twitch live stream.
Tesla has secured a two year special exemption in Norway to test its full self driving with supervision on public roads. This is another big step forward for Tesla and autonomous driving in Europe, as Norway is now the second country to allow this kind of testing in public.
Google’s DeepMind has developed an AI driven virtual fruit fly model that can simulate realistic walking, flying and general fly like stuff using a physics engine called MujoKo.