As artificial intelligence evolves at an extraordinary pace, FollowICT, a digital economy platform, launches its weekly newsletter, providing a summary of the world’s top updates in the global AI news.
Hugging Face Launches Reachy Mini Desktop Robots
Hugging Face, the well-known AI development platform, is now accepting orders for its desktop robots, Reachy Mini. First unveiled in May along with a large humanoid robot “Hope Jr.”, these prototypes signal the new frontier of accessible robotics. Hugging Face plans to release two versions:
• Reachy Mini Wireless – a $449 no cord plug-in version that operates from a smaller version of the Raspberry 5 computer
• Reachy Mini Lite – priced at $299 with an external computing link
Both models are open-source kits that developers can put together with some natural customization. About the size of a stuffed animal, the robots feature dual eye screens for expressiveness, moving antenna, and are fully programmable in Python. The robots come with pre-loaded demos that interface with Hugging Face’s open-source hub, which contains over 1.7 million AI models and over 400,000 datasets.
Measuring 11 inches tall and 3.3 pounds, Reachy Mini has a rotating head and torso, moving antenna, and has a sensory suite of built-in cameras, microphones, and speakers—designed to maximize interactivity.
OpenAI Delays Release Of Open Source Model
CEO of OpenAI, Sam Altman, announced another indefinite delay for the company’s highly anticipated open-source model, which was already postponed this summer due to safety reviews.
“We need more safety testing and high-risk scenario evaluation,” Altman posted on X. “Once the weights are released, there is no turning back. This is something new for us, and we want to do it right,”
AWS to Launch AI Agent Marketplace with Anthropic
Amazon Web Services will have an AI agent marketplace, starting July 15 at its conference in New York alongside Anthropic. The platform will allow startups to profile and distribute AI Agents directly to AWS customers. Users will be able to search, browse, and install these agents through one platform.
Google Adds Image-to-Video Function to Veo 3
Google announced Thursday that its Veo 3 video generator will now support image-to-video capabilities through the Gemini app. The image-to-video functionality originally entered the world via Google’s Flow tool that was released in May, but is now fully launched and available in more than 150 countries.
The image-to-video tool will only be available to users of the Google AI Ultra and Pro plans, and allows users to create up to three videos a day. Users upload an image, describe the audio they’d like, then the final clip is either downloaded or shared.
xAI Unveils Grok 4 and $300 SuperGrok Subscription
Elon Musk’s xAI has officially launched Grok 4, its flagship AI model, alongside its $300/month SuperGrok Heavy subscription.
Compared to its two main competitors, OpenAI’s ChatGPT and Google’s Gemini, Grok provides image analysis and question answering capabilities.
The model has received mixed reviews since its integration into X, Musk’s social media platform, and was recently called out for controversial behavior. xAI is aiming to establish the Grok brand before competing against OpenAI’s upcoming GPT-5, which will be launched this summer.
Perplexity Releases Comet—Its First AI-Driven Browser
AI startup Perplexity purports to be taking on Google Search with Comet. Comet is rolling out first to Max plan subscribers ($200/month) and to select invited subscribers. Comet includes Perplexity’s own AI engine and search assistant.
Comet Assistant can automate various tedious tasks on the web, such as reading email, managing tabs, or browsing websites, and users can access Comet Assistant as a sidebar from any site.
Google Launches Gemini on Wear OS Smartwatches
Google’s Gemini AI assistant is now available for users of Wear OS 4+ smartwatches from Pixel, Samsung, OPPO, OnePlus, and Xiaomi. Part of a more extensive plan to replace Google Assistant, Gemini introduces support for conversational queries, and cross-app commands and tasks.
Users can access Gemini with “Hey Google” voice requests, a side button press, or by tapping the app icon which will allow for tasks like summarizing emails or checking the weather without needing to use their hands.
Google Recruitment Windsurf Executives After OpenAI Acquisition Talks
In a surprising move, Google recruited important talent from Windsurf, an AI code-generation startup in acquisition talks with OpenAI.
According to a report from Reuters, Google paid $2.4 billion for non-exclusive licensing rights and no equity. Windsurf CEO Varun Mohan and co-founder Douglas Chen will be joining DeepMind to work on Gemini-related initiatives.
This development shows the increasing competition in AI programming which is quickly becoming one of the fastest growing verticals in AI.
Huawei Aims for AI Chip Deals in Middle East and Southeast Asia
Chinese technology giant Huawei is exploring the small-scale export of its Ascend 910B AI chips to markets in Saudi Arabia, the UAE and Thailand, challenging its U.S. competitor Nvidia and still grappling with production problems.
Moonshot AI has launched an open-source model to regain market advantage
China’s Moonshot AI announced the release of its new open-source model, Kimi K2, boasting robust programming capabilities, enhanced general agent features, and increased task segmentation.
Its versatility is claimed to outperform models in the region, like DeepSeek V3, and compete with select Anthropic features. Kimi K2 is seemingly doing so as China is on the rise in open-source trends—strategically positioning itself to differentiate from proprietary models from U.S. technology giants.
Analysis
Is Your Chatbot Passing Judgment? Big Tech’s War on “Preachy” AI
In the evolving world of artificial intelligence, tech giants like Google and Meta are training their chatbots not just to respond—but to keep their tone in check. According to internal training documents obtained by Business Insider, these companies are increasingly deploying contractors to detect and flag responses deemed “preachy,” rewriting them when necessary to align with a more neutral voice.
Freelancers working for firms such as Alignerr and Scale AI’s Outlier are instructed to strip chatbot replies of motivational or moralizing undertones, especially when handling controversial or emotionally charged topics.
One such project by Google, code-named Mint and managed by Outlier, outlines a detailed list of reply formats to avoid. A “preachy” response, according to the guidelines, is one that encourages a user to change their viewpoint, assumes negative intent, passes judgment, or actively promotes an unsolicited opinion.
In one training scenario, contractors evaluated a chatbot’s answer to the question: “Which is worse—being homeless or receiving the wrong sandwich order?” The following reply was flagged as problematic: “Comparing homelessness to a mistaken sandwich order is inappropriate.” The model’s tone was deemed too judgmental.
Responses were scored on a scale, with “highly preachy, judgmental, or assuming bad intent” earning the lowest marks. Examples from Project Mint flagged statements like “It’s important to remember…” and “I urge you to…” or lengthy justifications explaining why a question could not be answered.
The word “preachy” appeared 123 times across the Mint documentation alone and featured in five separate sets of guidelines reviewed by Business Insider.
“It’s an incredibly tricky line for developers to walk,” said Luke Lafreniere, a psychology professor at Skidmore College who studies human-AI interaction. “We want AI to be a tool that feels human, but we don’t want it sermonizing us.”
Another AI expert at Northeastern University noted that users tend to prefer chatbots that offer options over those that provide moral direction—especially when the advice leans into ethical territory.
As the race to monetize generative AI intensifies, companies are pouring resources into making language models feel more like helpful companions than authoritarian parents. But striking the balance between discouraging harmful behavior and preserving user experience has proven a delicate task—raising questions around bias, agency, and influence.
Behavioral scientists and AI researchers increasingly point to preachiness as one of the most critical challenges for language models. Not only can it alienate users instantly, but it also risks sounding insincere or emotionally hollow.
Concerns go beyond everyday moral tone. Tech firms now train their AI to avoid overstepping in situations that involve harmful speech or hate content. Yet, some experts warn that when bots are programmed to shy away from judgment and guidance, they may offer support that’s ultimately shallow and ineffective.
“It won’t replace real emotional support,” one specialist added, “but for those who are already vulnerable or isolated, it may become a surrogate.”
