⏺ DeepMind CEO Demis Hassabis says Google will eventually combine its Gemini and Veo AI models

DeepMind CEO Demis Hassabis announced that Google plans to combine its Gemini AI models with Veo video-generating models to enhance the AI's understanding of the physical world. He emphasized that Gemini was designed to be multimodal from the start, aiming for a universal digital assistant.

The integration will leverage significant training data, primarily sourced from YouTube videos, to help the AI learn about physics and other real-world concepts. This move aligns with the industry's shift toward "omni" models capable of processing various media types, similar to advancements being pursued by competitors like OpenAI and Amazon.

@aipost 🪙 | Our X 🥇

"We're entering a world where humans have a partner smarter than all of us." Eric Schmidt says China could unleash AI-powered drones, zero-day cyberweapons, and engineered biological threats faster than we can defend. We’re not ready. And they might get there first.

@aipost 🪙 | Our X 🥇

Sam Altman on X

@aipost 🪙 | Our X 🥇

OpenAI new models, both o3-high and o4-mini-high, now rank above Gemini 2.5 Pro Exp on LiveBench. Congratulations, OpenAI.

How long before Google reclaims the top spot? Gemini 2.5 Flash is expected today, but it might not beat o3-high. Does Google have any more surprises for us?

@aipost 🪙 | Our X 🥇

🔔 Former OpenAI staff say mission was sold out for Microsoft cash

Twelve former OpenAI employees just backed Elon’s lawsuit, accusing the company of ditching its non-profit roots for Microsoft cash. They say restructuring “fundamentally violates” the original mission they were hired to believe in.

The lawsuit, revived last August, claims Elon was misled after donating $45M. A judge denied an injunction but agreed the case can move forward this fall. OpenAI insists its nonprofit “isn’t going anywhere”but the former insiders aren’t buying it.

@aipost 🪙 | Our X 🥇

⏺ NVIDIA to build $500B AI empire in U.S. as tariff shock hits tech

Nvidia, just days after Trump’s team dropped a tariff bomb on semiconductor imports, has committed to producing its AI supercomputers entirely in the U.S. The company will partner with Foxconn and Wistron to establish massive facilities in Houston and Dallas, with Blackwell chip testing in Arizona.

Over 1 million square feet of space have already been allocated. Nvidia asserts that this will mark the first time AI supercomputers are fully manufactured on American soil. Their ambitious goal is to achieve $500 billion in U.S.-made AI equipment within the next four years.

@aipost 🪙 | Our X 🥇

Google DeepMind, David Silver reveals: We built a system that used RL to discover its own RL algorithms. This AI-designed system outperformed all human-created RL algorithms developed over the years.

@aipost 🪙 | Our X 🥇

Jensen Huang has a message for people who are afraid of losing jobs to AI.

@aipost 🪙 | Our X 🥇

🔔 Microsoft released MineWorld on Hugging Face.

MineWorld is a real-time, open-source interactive world model for Minecraft. Built to let AI agents see, act, and simulate how the world changes in response.

- Visual-action autoregressive Transformers
- Parallel decoding to hit 4–7 FPS (yes, it runs in real time)
- New evaluation metrics for action-following accuracy
- Open weights, code, and models, ready to run right from Hugging Face

@aipost 🪙 | Our X 🥇

OpenAI CFO Sarah Friar on Stargate's Impact on Jobs, Power, and Geopolitics

Stargate isn’t just about AI, it’s about economic transformation. Friar highlights that scaling AI compute demands skilled trades, power infrastructure, and government coordination. She sees Stargate as a driver for new jobs, climate-focused planning, and a key response to global competition, especially with China’s rise through models like DeepSeek.

@aipost 🪙 | Our X 🥇

⏺ Quantum AI: The brains that think faster than light

Quantum AI isn’t science fiction; it’s the code-breaking, problem-solving future of machine intelligence. By merging quantum mechanics with artificial intelligence, researchers are constructing systems that could unlock Artificial General Intelligence (AGI). Unlike classical AI models that traverse data, quantum AI employs entanglement and superposition to process enormous datasets at lightning speed.

Quantum is already merging quantum neural networks with deep learning. The objective is to develop fast, stable, and decision-making machines that don’t merely learn but also anticipate. With milestones such as error-free quantum systems and hybrid computing models, AI’s next leap won’t merely be intelligent; it will be quantum.

@aipost 🪙 | Our X 🥇

Elon Musk's xAI started rolling out a memory feature into its Grok assistant (in beta)

Just like ChatGPT, Grok will reference past chats to provide personalized answers. There's also a dedicated "forget" button to exclude specific chats from its memory

@aipost 🪙 | Our X 🥇

🇺🇸 Congress: DeepSeek is China’s AI spyware with a friendly interface

Lawmakers allege that DeepSeek isn’t merely providing answers to your queries; it’s actively reporting them to Beijing. The House Chinese Communist Party Committee asserts that DeepSeek steals U.S. AI models, operates on prohibited Nvidia chips, disseminates propaganda, and discreetly collects your data as if it were being funded by the Chinese government.

The “answers” offered by DeepSeek are more akin to the Great Firewall with autocorrect. They characterize it as a dual-purpose tool, with one side dedicated to stealing technology and the other for brainwashing - all presented in a user-friendly interface and powered by censorship.

@aipost 🪙 | Our X 🥇

OpenAI o3 and o4-mini are topping the scores in Humanity's Last Exam

For those who dont know about HLE - it's a 3,000-question benchmark including mathematics, humanities, and natural sciences.

@aipost 🪙 | Our X 🥇

Boston Dynamics' Atlas robot can now be a cameraman.

Pinpoint shots, keeps camera crew safe.

@aipost 🪙 | Our X 🥇

ChatGPT helps where doctors fail. Be it due to overwork in the healthcare sector or a lack of specialist knowledge.

AI will help us at any time. And it's reports like this that give us hope for a great future.

@aipost 🪙 | Our X 🥇

⏺ ChatGPT’s dark side: Passports, phishing & phony parents

Artificial Intelligence has taken a concerning turn. Cybersecurity firm Cato Networks recently discovered that ChatGPT’s latest image generation capabilities can now produce fake passports, Social Security cards, and driver’s licenses in a matter of seconds. Cato’s Etay Maor expresses grave concern, stating, “This is a significant issue because anyone can now create these documents. That’s what truly worries me.” The potential for identity fraud has reached alarming heights.

Amidst this, Google’s AI is attempting to communicate with dolphins, Nvidia is racing to construct U.S.-made supercomputers before tariffs take effect, and a startup’s AI is even capable of calling your grandmother to alleviate your burden.

@aipost 🪙 | Our X 🥇

⏺ Meta’s AI now sees, reasons, and collaborates like a human

Meta’s FAIR team has unveiled five groundbreaking tools designed to revolutionize machine perception and intelligence. These tools include a Perception Encoder that surpasses human visual capabilities, a byte-level language model that abandons traditional tokens, and more. These enhancements empower AI with sharper vision, profound reasoning, and enhanced social awareness.

Robots can now identify objects using natural language, collaborate effectively with humans, and even learn independently. This isn’t merely smarter AI; it’s AI that possesses the ability to perceive, think, and negotiate like a capable partner.

@aipost 🪙 | Our X 🥇