The artificial intelligence (AI) realm witnesses rapid developments around the world, reshaping the world of digital transformation and accelerating the pace of AI adoption. In this weekly newsletter, FollowICT is highlighting the most important news and advancements in AI around the globe.
Watch the Video: Germany Unveils Human Robot 4NE-1 to Challenge China and America’s Dominance
Neura Robotics, a European robotics company, plans to release its third-generation human robot, the 4NE-1, in June. The company asserts that it will be the best on the market. The German firm specializes in developing advanced robots with sensory capabilities.
With the rapid progress in robotics and growing geopolitical competition, especially between Europe and the United States, this launch highlights Europe’s push to maintain its competitive edge in the expanding automation sector. In January 2025, Neura Robotics raised $120 million in funding aimed at accelerating its efforts to develop “cognitive” robots with enhanced cognitive and data processing abilities.
Last year, Neura Robotics revealed details about the third-generation human robot, 4NE-1, which is set for release this June.
The robot stands 5.9 feet tall (1.8 meters) and weighs 176 pounds (80 kg), capable of carrying loads up to 33 pounds (15 kg). The 4NE-1 is designed to handle a variety of daily and industrial tasks and operates on Neura’s Neuraverse platform with a top speed of 1.8 mph (3 km/h). Its standout feature is its interchangeable arms, which can be customized for different tasks.
Google Enhances Vertex AI with Chirp 3 for Voice AI
Google has announced the addition of its Chirp 3 voice model to its Vertex AI platform, paving the way for more advanced speech-to-text and text-to-speech technologies.
While most generative AI research has focused on text and images, Google is making a significant push into voice technology, revealing that Chirp 3 will be supported on its cloud platform starting next week.
This follows Google’s announcement of 8 new voices supporting 31 languages, boosting the capabilities of voice assistants, audiobooks, customer support agents, and video voiceovers.
1X Tests Human-like Robots in ‘Several Hundred’ Homes This Year
Norwegian robotics company 1X plans to begin early testing of its human-like robot, Neo Gamma, in “several hundred to a few thousand” homes by the end of 2025, according to the company’s CEO, Bernt Bornish, as reported by TechCrunch.
Bornish said, “The Neo Gamma system will enter homes this year. We want to invite early users this year to help us develop this system. We want it to live and learn among people, and to achieve that, we need people to bring Neo into their homes and help us teach it how to behave.”
– Anthropic Adds Web Search to its Claude Chatbot
Anthropic’s AI-powered chatbot, Claude, can now search the web—a long-awaited feature.
Anthropic announced on its blog that the web search feature is now in beta for paid Claude app users in the U.S., with plans to expand support for free users and other countries soon. Users can activate the web search feature in their profile settings within the Claude web app, and Claude will automatically search various sites to provide specific answers.
Currently, the web search feature is only available with Anthropic’s latest model, Claude 3.7 Sonnet.
SoftBank Acquires Ampere Computing for $6.5 BLN
SoftBank announced on Wednesday that it has acquired Ampere Computing, a Silicon Valley-based chip design startup, for $6.5 billion in cash.
SoftBank stated that Ampere will operate as a subsidiary but will retain its name and headquarters in Santa Clara, California.
Masayoshi Son, CEO of SoftBank, said in a statement, “The future of advanced AI requires exceptional computing power. Ampere’s expertise in semiconductors and high-performance computing will help accelerate the realization of this vision, further cementing our commitment to AI innovation in the U.S.”
Roblox Launches AI Technology for 3D Object Drawing
Roblox, the platform for developing video games and other products, has announced the launch of “Cube 3D,” its first AI model for creating 3D drawings of objects.
Game developers will be able to access Cube 3D later this week, and the company has also released an open-source version of the model.
Cube 3D allows users to create 3D models of objects with a single click. For example, users can “create an orange race car with black stripes.”
The open-source version of the model lets anyone customize it, create additional components, or train it using their own datasets to suit specific needs.
New Enhancements to ChatGPT’s Image Creation Capabilities
OpenAI’s chatbot is about to get improvements that will strengthen its position in the ongoing competition with Google’s “Gemini.”
Gemini stands out by being pre-installed on most flagship Android phones, but ChatGPT often includes newer features and models that help it excel.
ChatGPT already supports image generation in its Android app, but there may soon be an update enhancing this capability. The release code for “v1.2025.077” of ChatGPT for Android suggests improvements to the chatbot’s image creation abilities are on the way.
Nvidia and Google DeepMind Help Power Disney’s Cute Robots
Nvidia CEO Jensen Huang announced at the GTC 2025 conference on Tuesday that the company is collaborating with Disney Research and Google DeepMind to develop Newton, a physics engine designed to simulate robotic movements in real-world environments.
Disney will be among the first companies to use Newton to operate next-generation entertainment robots, such as the BDX droids inspired by Star Wars—one of which was showcased alongside Huang during his keynote on Tuesday.
In a press release, Kyle Laughlin, Senior Vice President of Disney Imagineering, stated that the collaboration with Nvidia and Google DeepMind will play a key role in powering Disney’s future entertainment robots as well.
Newton is expected to help robots become more “expressive” and “learn to perform complex tasks with greater precision,” according to Nvidia. The physics engine is designed to help developers simulate how robots interact with the natural world, which can sometimes be a challenge for robotics developers.
Musk and Microsoft Partner in AI Race with $30 BLN Project
Elon Musk continues to surprise everyone, this time by teaming up with Microsoft in a massive $30 billion project to develop data centers.
Microsoft, alongside BlackRock and UAE-based MGX, announced a new partnership with Musk’s emerging company, xAI, in a move that could reshape the competition landscape in artificial intelligence.
Microsoft, the biggest backer of OpenAI, has decided to collaborate with its new competitor as it seeks to reduce its reliance on OpenAI and develop its own AI models, such as MAI.
The exact financial details of xAI’s investment in the project have not been disclosed, but the startup has raised $12 billion since its founding in 2023 and is looking to raise an additional $10 billion.
Reports indicate that xAI’s “Colossus” data center in the U.S. will be the largest of its kind, housing over a million GPU units to run the “Grok” chatbot.
Sam Altman Advises Students to Master AI Tools as They Once Did Programming
CEO of Open AI Sam Altman offers advice to students preparing to enter a job market increasingly reliant on programming automation: master AI tools.
Altman said, “The tactical thing is to master using AI tools. Just like when I was a senior in high school, the tactical thing was to master programming. This is the new version of that.”
He added, “I think the percentage of programming being done by AI in many companies is now over 50%. But I think the big breakthrough is going to come with agent-based programming, which no one has really done yet.”
When asked why this hasn’t been achieved, Altman responded, “We just need a little more time,” noting that the problem lies with the model, not the product.
For students preparing for careers, Altman advised them to develop “adaptability and learning ability” rather than learning specific skills.
xAI Launches Image Generation API
Elon Musk’s AI company, xAI, has added image generation capabilities to its API.
Currently, there is only one model available in the API, “grok-2-image-1212.” With an accompanying caption, the model can generate up to 10 images per request (with a maximum of five requests per second) in JPG format, priced at $0.07 per image.
Baidu Launches Two New Free AI Models
Chinese tech giant Baidu has launched two new free AI models as part of its efforts to regain its leading position in the increasingly competitive AI landscape in China.
The models, released on Sunday, include the company’s first model focused on reasoning, ahead of plans to transition to an open-source model strategy.
Baidu said its new reasoning model, “ERNIE X1,” offers performance on par with DeepSeek’s R1 model at half the price, and features “stronger capabilities in understanding, planning, reasoning, and development.”
AI Tool Generates High-Quality Images Faster Than Other Models
The ability to generate high-quality images quickly is crucial for creating realistic simulation environments used to train self-driving cars to avoid unforeseen hazards, making them safer on real streets.
However, generative AI technologies, which are increasingly used to produce such images, have their drawbacks. One popular model, the diffusion model, can produce stunningly realistic images, but it is very slow and requires intensive computations for many applications. On the other hand, self-regressive models, which power LLMs like ChatGPT, are much faster but often produce lower-quality images full of errors.
Researchers from MIT and NVIDIA have developed a new approach that combines the best of both methods. The hybrid image generation tool first uses a self-regressive model to quickly capture the full image, then applies a small diffusion model to enhance the image details.
Known as HART (Hybrid Automatic Regression Transformer), the tool can create images that match or exceed the quality of modern diffusion models but does so roughly nine times faster.
The process consumes fewer computational resources than traditional diffusion models, enabling HART to run locally on a laptop or commercial smartphone. A single natural language prompt in the HART interface is sufficient to generate an image.
HART has a wide range of applications, including helping researchers train robots to complete complex real-world tasks and assisting designers in creating stunning visuals for video games.
OpenAI’s “o1-pro” Becomes the Most Expensive AI Model
OpenAI has launched a more powerful version of its “reasoning” AI model, o1, called “o1-pro,” in its developer API.
According to OpenAI, o1-pro uses greater computing power than o1 to consistently provide better responses. OpenAI charges $150 per million tokens (about 750,000 words count) input into the model, and $600 per million tokens generated by the model.
OpenAI bets that the enhanced performance of o1-pro will convince developers to pay these hefty fees.
NVIDIA Reveals Human-like Robot Models at GTC
NVIDIA announced the Groot N1 model at the GTC 2025 conference in San Jose, a general-purpose model trained on both synthetic and real data.
In an introductory video for Groot N1, NVIDIA mentions that it features a dual-system architecture for fast and slow thinking, inspired by human cognitive processes.
Groot N1 is an evolution of NVIDIA’s Groot project, which the company introduced at last year’s GTC.
The Groot project was initially aimed at industrial use, but Groot N1 expanded its scope to include human-like robots across a range of form factors.
Meet 5 Chinese AI Models That Are Revolutionizing the Market
The AI scene in China is heating up, with new models emerging to compete with the best players in the US and shaking up the stock market. As Chinese AI models gain ground rapidly, here are five key players to watch.
- DeepSeek’s R1
DeepSeek was the first Chinese AI startup to make a breakthrough in the field. In January, its new model caused a sharp drop in U.S. tech stocks and raised questions about the decline of U.S. leadership in AI.
The startup, based in Hangzhou, was founded in 2023 by hedge fund entrepreneur Liang Wenfeng.
After purchasing thousands of Nvidia chips, Wenfeng built DeepSeek’s R1 thinking model on its V3 core model, training it on around 2,000 Nvidia H800 chips at a total cost of approximately $5.6 million, according to the startup.
Despite its lower cost, the R1 model competes with major players like ChatGPT’s o1, according to DeepSeek.
Its impact on the AI chip market was immediate: NVIDIA’s stock dropped by about 17%, wiping out hundreds of billions of dollars in market value, according to Business Insider.
- Alibaba’s QwQ-32B Model
Alibaba has become a major player to watch after it released the open-source QwQ-32B model in early March. The company stated that this model used fewer data compared to DeepSeek’s.
Alibaba said its QwQ-32B model, which has five parameters of DeepSeek’s R1, is designed for efficiency.
The model is now open-source on platforms like Hugging Face, allowing free sharing of the code with anyone for any purpose.
- Tencent’s Yuanbao
Tencent’s chatbot, Yuanbao, topped China’s iOS App Store in early March, surpassing DeepSeek’s app as the most downloaded free app, according to a Bloomberg report.
Tencent, based in Shenzhen, operates WeChat, China’s largest social media app, used by nearly 1.4 billion people.
Last month, the South China Morning Post reported that Tencent added a download button for Yuanbao to WeChat, making the chatbot more accessible to app users.
Yuanbao is now the third most downloaded free app on China’s iOS App Store, ahead of DeepSeek and ByteDance’s Doubao. Doubao is a chatbot similar to ChatGPT.
- Manus
Manus is the latest AI innovation from China. According to multiple reports, it was designed by the Chinese startup Monica, which seems to be affiliated with the company Butterfly Effect. The privacy policy of Manus states that Butterfly Effect is a registered entity in Singapore.
Researchers at Monica confirmed that Manus is the first fully independent AI client in the world. Unlike chatbots that require multiple inputs, it can perform complex tasks like sorting resumes, analyzing stocks, gathering data, and even creating websites with just a single prompt.
According to Manus’s website, it outperforms OpenAI’s deep search model in the GAIA benchmark, a tool for comparing models.
- Baidu’s Ernie X1 Model
This month, Baidu, China’s competitor to Google, launched two new versions of its AI model, Ernie.
Ernie X1 is an inference model that Baidu claims can “deliver performance comparable to DeepSeek R1 at only half the price.”
Baidu added that the model has “stronger capabilities for understanding, planning, reflection, and evolution.”
Ernie 4.5 is an updated version of the company’s core model, which Baidu says “outperforms GPT-4.5 in multiple benchmarks while costing only 1% of GPT-4.5’s price.”
With the release of these updated models, Baidu stated that it will integrate Ernie X1 and Ernie 4.5 into its search engine and broader product suite.
