👁️ 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 🏴

⚡️ AI is now doing your grocery runs.

Gopuff launched an AI shopping assistant called "Go," built with SpaceXAI. You say what you need. It handles the rest.

@aipost 🏴

What does the lead artist behind Telegram Gifts think about AI? 🗣

📄 In a new interview with Just News, Ohuenko shares how AI has become one of his primary tools for visualizing ideas — and why he believes it will eventually outperform humans at almost everything.

⚡️His conclusion is simple: in the age of AI, the only reason to keep creating is to genuinely love the process.

The @just interview also covers Telegram Gifts, creativity, stickers, digital art, and the future of content on Telegram. 🐸

@aipost 🏴

👁️ NVIDIA has launched new physical AI and robotics development skills designed to automate key workflows from simulation and scene preparation to training and deployment.

@aipost 🏴

Robot unboxing to soon be a familiar sight in households

Robot unboxing scenes are expected to become a common occurrence in homes worldwide. This development may arrive sooner than many people anticipate.

Industry trends indicate a rise in household adoption of robotics, with widespread availability projected in the near future.

📰 @aipost

🇦🇷 Argentina wants to give AI its own companies

While much of the world is debating how to regulate AI, Argentine President Javier Milei is proposing something far more radical.

In a Financial Times column, Milei unveiled a plan to turn Argentina into a global AI hub built on three ideas:

• Keep AI largely free from regulation.
• Create “non-human corporations” that can be run by AI agents or robots.
• Offer low taxes and business-friendly rules to attract founders and investors.

The most surprising proposal is the second one.

Under the plan, AI-controlled organizations could gain legal recognition and limited liability protections, allowing autonomous agents to operate businesses in their own right.

Milei compares this moment to the birth of the modern corporation during the age of global trade, arguing that new legal structures are needed for the AI era.

@aipost 🏴

❗️Ray Kurzweil still thinks human-level AI arrives by 2029

Back in 1999, futurist Ray Kurzweil predicted that AI would reach human-level intelligence by 2029. At the time, many top researchers at Stanford University thought it was at least 100 years away. Now, with just three years left, Kurzweil says only two major pieces are still missing:

• AI that truly understands the physical world.
• Robots that can reliably navigate everyday common sense like knowing what belongs in the fridge and what belongs in the trash.

His argument is simple: progress looks slow until it suddenly doesn’t.

The Human Genome Project was only 1% complete halfway through. One final doubling later, it was finished. Kurzweil believes AI is approaching a similar moment.

Just a year ago, he says large language models were interesting but not especially impressive. Today, AI is writing code, conducting research, and in some medical studies outperforming human doctors in diagnosis.

Meanwhile, computing power has exploded.

Over the last 70 years, hardware performance improved by roughly 75 quadrillion-fold, while software efficiency improved by around a million-fold.

Kurzweil’s bet is that when exponential curves become visible, they stop looking like curves and start looking like walls. And according to him, we’re getting very close to that wall.

@aipost 🏴