Highlights from Satya Nadella’s keynote at Microsoft Build.

@aipost 🏴

🔥Microsoft unveils MAI-Voice-2, pushing AI voices closer to human speech

Microsoft has introduced Microsoft MAI-Voice-2 and MAI-Voice-2-Flash, its latest text-to-speech models designed to generate more natural, expressive, and multilingual voices.

🔹MAI-Voice-2

• Focused on high-quality speech generation.
• Better emotional expression and voice control.
• More consistent long-form narration for audiobooks, podcasts, and assistants.
• Supports multiple languages and speaking styles

⚡ MAI-Voice-2-Flash

• Optimized for low-latency, real-time applications
• Built for AI agents, voice assistants, and large-scale deployments
• Designed to handle high-volume workloads while maintaining natural speech

The launch reflects a broader industry trend: AI voice models are rapidly evolving from simple text readers into systems that can convey tone, emotion, and personality almost like a human speaker.

Source.

@aipost 🏴

Microsoft launches Scout, an AI executive assistant that shows up directly in your email and calendar like a real employee.

Source.

@aipost 🏴

China has introduced a virtual reality shooting experience that transforms shopping malls into interactive play areas.

Participants use gun controllers equipped with smartphones, allowing for real-time augmented reality overlays. The system projects virtual enemies and effects directly onto the actual mall environment through the device’s camera feed.

This technology blends digital gaming elements with physical locations, creating a mixed-reality environment within public spaces.

📰 @aipost

Microrobots aid recovery in mice with spinal cord injuries

Researchers developed six-micrometre stem-cell microrobots, guided by magnetic fields, to help recover movement in mice with severed spinal cords. These microrobots transformed stem cells into neurons through wireless stimulation, eliminating the need for implanted electrodes.

After 28 days, connected nerve cells improved movement in the mice, with no adverse effects noted. A similar approach in zebrafish showed quicker recovery, with near-normal swimming in just three days.

These scalable microrobots can be produced on a lab-on-a-chip within thirty minutes and dissolve after use, with potential applications in areas like cardiology and wound healing.

📰 @aipost

🤖 Anthropic is giving a select group access to one of the most powerful cyber-defense AI models ever built.

The company is expanding Claude Mythos Preview from around 50 to 200 vetted organizations across critical sectors including healthcare, energy, water, communications, governments, nonprofits, and major software projects.

The expansion includes the US-led military alliance NATO, the US tech group Okta, the EU cybersecurity agency ENISA, and South Korean companies Samsung, SK Hynix and SK Telecom.

Source.

@aipost 🏴

🇺🇸 Marco Rubio warns Americans to prepare now for a future where artificial intelligence rapidly replaces millions of jobs, forcing workers to learn new skills or risk being left behind.

Rubio says the AI revolution won't just threaten blue-collar work, warning that millions of white-collar professionals could soon find themselves replaced by machines.

"There will be white-collar jobs that will be impacted."

"That over time could destabilize societies all over the world."

@aipost 🏴

❗️Bernie Sanders announces he will soon introduce the American AI Sovereign Wealth Fund Act.

It proposes a one-time 50% tax on stock (not profits) in America’s largest AI companies, giving the public a direct 50% ownership stake.

@aipost 🏴

🤣🤣

@aipost 🏴

🗣️ AI may be changing the way humans speak.

According to Harvard-trained linguist Adam Aleksic, one of the clearest signs is a surprisingly ordinary word: “delve.”

Since ChatGPT launched, usage of the word has reportedly surged by around 1,000%.

Why? Aleksic argues it’s partly a result of how AI models are trained. Human reviewers in regions where words like “delve” are somewhat more common helped shape model outputs, while the models themselves tend to favor more formal, Latin-derived vocabulary that sounds authoritative.

The result: AI frequently uses words like delve, and people are starting to copy that style.

What’s fascinating is that the influence isn’t staying inside the machine. Evidence suggests humans are increasingly using AI-favored words in everyday speech, creating a feedback loop where AI shapes the language of the people who trained it.

As podcaster Chris Williamson put it:

“The creature that programmed the AI is being programmed by the AI.”

Aleksic’s response:

“Its reality is influencing our reality.”

Language has always evolved through books, television, and the internet. Now, AI may be becoming the newest force shaping how millions of people communicate.

@aipost 🏴

❗️AI bots have officially overtaken humans on the internet

Cloudflare CEO Matthew Prince shared new data showing that bots and AI agents are now generating more internet traffic than humans for the first time in history.

@aipost 🏴

👁️ NVIDIA Introduces Nemotron 3 Ultra.

A frontier smart open model built for long-running agents that need to plan, reason, use tools and keep working across complex coding, research and enterprise workflows.

Source

@aipost 🏴

🔍 Google introduces Gemma 4 12B, a multimodal open model that runs on a laptop

Google has unveiled Gemma 4 12B, its latest open-weight model with advanced reasoning, vision, and audio capabilities. Despite its size, it delivers performance close to larger Gemma models while running locally on just 16GB of VRAM and is released under the permissive Apache 2.0 license.

The most interesting part is its new unified architecture.

Most multimodal models rely on separate vision and audio encoders, which add memory overhead and latency. Gemma 4 12B largely removes them:

👁 Vision: A lightweight embedding module replaces the traditional vision encoder, allowing the LLM itself to handle visual understanding.

🎤 Audio: The audio encoder is removed entirely. Raw audio is projected directly into the same token space as text.

The result is a smaller, faster, and more efficient multimodal model that can run on consumer hardware without sacrificing much capability.

This could be a glimpse of where AI architectures are heading: fewer specialized components, more native multimodal reasoning. 🚀

Source.

@aipost 🏴