lunin

nostr

*Today is the first anniversary of Nostria, a social network app I have built, which I have poured my energy and soul into. It's taken a lot of effort to get here and the people who have started using it, is enjoying it a lot, I think you will too. Now I need your help with one small thing: Please support Nostria on ProductHunt! Thank you all!* - ## [Sondre Bjellås](https://www.linkedin.com/in/sondreb/)

Support "Nostria - premium Nostr infrastructure" on Product Hunt today!

Runway demonstrated real-time video generation.
The new model creates HD video instantly. The latency to the first frame is less than 100 ms.
It runs on the latest Nvidia Vera Rubin chips, which will be in mass production in the second half of the year. The project itself is currently a research preview.

A breakthrough in real-time video generation

**NVIDIA held the main AI conference of the year in San Jose.**
Six things you need to know:

**Vera Rubin.** New architecture: 7 chips, 5 rack-mount systems. 10x the performance per watt of Blackwell. Shipments by the end of 2026.

**Groq 3.** In December 2025, NVIDIA acquired the technology of the startup Groq for $20 billion. Groq was making chips for running models, not training them. Groq 3 LPU: the first product from this deal. The market is good at training models, but running them quickly and cheaply for millions of users is a different challenge. NVIDIA is now closing both.

**$1 trillion.** Huang stated that combined orders for Blackwell and Vera Rubin will reach $1 trillion by 2027. A year ago, the forecast was $500 billion. That's a doubling in 12 months.

**Robotaxis**. NVIDIA has partnered with Uber to launch robtaxi services in 28 cities on four continents by 2028. BYD, Hyundai, Nissan, and Geely are building Level 4 cars on the NVIDIA Drive platform. Level 4 means the car drives itself, no driver needed.

**Space.** Space-1 Vera Rubin: AI data centers that will operate in orbit. The idea is to process data from satellites directly in space, without wasting time and bandwidth transmitting it to Earth.

**Feynman.** NVIDIA announced a generation AFTER Vera Rubin, which hasn't even been released yet.

GTC 2026: What Huang Showed in Two Hours on Stage

What's unique about it is that training is conducted in-house from scratch, without RAGs, without fine-tuning existing LLMs, and without dependence on the model provider's API.

Mistral's open-source models can be used as a basis.

Mistral launched Forge, a platform for corporate LLM training

Nvidia's AI agent platform is called NemoClaw. It's essentially a modified OpenClaw with a focus on secure operation for businesses.

You can connect any agent, cloud or on-premises. It's not tied to Nvidia chips—you can use any hardware.

The alpha version is now available.

Nvidia Announces "Secure OpenClaw"

Mini has improved coding, reasoning, multimodal understanding, and tool usage. Nano is the smallest and fastest GPT‑5.4. Recommended for subagents performing simple tasks.

GPT‑5.4 mini and nano released

Manus for desktop is now available.

Like OpenClaw, Manus offers the ability to install itself on "any old Mac Mini" to turn it into an AI agent that can be tasked 24/7 for autonomous execution.

Works on macOS and Windows.

Introducing My Computer: When Manus Meets Your Desktop

OpenAI is refocusing on corporations. Altman and his team decided that the "do it all" strategy had put OpenAI on the defensive against competitors. Side projects are being phased out. Programming and business products are now the priority.

OpenAI to Cut Back on Side Projects in Push to ‘Nail’ Core Business

The company has opened a series of job openings for financial experts. They will teach Grok financial transactions, including leveraged loan syndications, distressed asset investments, and more.

The terms are not public, but last year OpenAI offered $150 per hour for this type of work.

Musk’s xAI Hiring Credit Experts, Bankers to Teach Grok Finance

Nvidia unveiled DLSS 5 - AI for improved image quality in games.

The neural network recognizes objects in the frame in real time and applies photorealistic lighting and shadows to them.

Releasing in fall 2026, exclusively for RTX 50-series graphics cards.

NVIDIA DLSS 5 Delivers AI-Powered Breakthrough in Visual Fidelity for Games

A historic shift is underway in the LLM industry. Neural networks have learned to physically calculate.

The startup Percepta has published a study (currently a Proof of Concept) that has the entire AI community buzzing. They've baked a virtual machine (WASM interpreter) directly into the Transformer's weight matrix.

What's the fundamental problem?
Regular ChatGPT and Claude don't calculate math; they guess it (hence the hallucinations in exact problems). Currently, they're fixing this with hacks: they're forcing the AI to write Python code and run it in an external sandbox.

Percepta has proven that the Transformer is capable of executing the most complex machine code directly within itself with 100% accuracy!

What it looks like (pure cyberpunk):
Instead of generating text across the screen at 30,000+ tokens per second, machine code flies – the neural network juggles registers and branching. The authors fed the world's most complex Sudoku into the weights. The Transformer physically executes the search algorithm: it substitutes a digit, catches a contradiction, backtracks, and finds the answer. No hallucinations, pure logic.

How did they break the limitations?
They didn't do this before because of Attention Bottleneck.

A classic neural network is forced to reread its entire generation history at every step. At the millionth computational step, it would simply consume all the memory and die.
The authors invented Exponentially Fast Attention – searching past data in logarithmic time. The Transformer stopped being slow and started running millions of steps in seconds.

A direct path to AGI? Andrey Karpaty has already commented on the article, expressing his respect. And the point here is far from just creating a fast calculator.

Modern LLMs are our intuitive, "fast" thinking (System 1). Program code is rigid, deterministic logic (System 2). What Percepta has created is a blueprint for merging them into a single "brain."

If this infallible mathematical coprocessor can be seamlessly combined with speech models, we'll get AI that will forever stop hallucinating in logic. Agents will no longer need external scripts - heavyweight simulations will run right inside the scales. And this isn't just an improvement on chatbots; it's a direct, technically tangible bridge to AGI.

Can LLMs Be Computers? | Percepta

Now you can find out when AI will take your job - the first automation risk map for 342 professions has been released.

The author analyzed data on the US labor market, including worker education requirements, average salaries, and job responsibilities, and turned it all into an interactive map.

342 occupations · area = employment · color = AI exposure
Data from BLS, scored by Gemini Flash.

AI Exposure of the US Job Market

ByteDance has postponed the global release of its Seedance 2.0 video generator.

Seedance 2.0 is notable for its ability to create high-quality videos featuring Hollywood stars. This has probably prompted complaints from everyone in Hollywood about ByteDance.

For now, Seedance 2.0 will remain available only in China.

ByteDance reportedly pauses global launch of its Seedance 2.0 video generator

The AI analyzes medical records, test results, and wearable device readings. Based on all this, users will receive health recommendations and tips on what to ask their doctor at their next appointment.

It's essentially similar to ChatGPT Health, announced (https://t.me/aioftheday/4010) in January. Microsoft cites its advantage as being a "trusted brand," meaning people are more likely to share their secrets with it than with just anyone.

The beta version will be free, but currently only available in the US and by invitation. Access will become paid later.

Microsoft unveiled Copilot Health for health advice

Google has released a multimodal embedding model. Gemini Embedding 2 works with text, images, video, audio, and PDF. It outperforms all competitors in benchmarks. However, it appears there are no other models that cover this range of tasks.

Gemini Embedding 2: Our first natively multimodal embedding model

ChatGPT has started displaying interactive explanations of formulas in replies.

Over 70 math and science topics are supported. Currently, it only works in English.

New ways to learn math and science in ChatGPT

Video generation will soon move from a separate app directly into the chat.

The Information's sources believe that Altman is attempting to boost ChatGPT's popularity amid increased competition from other chatbots. OpenAI's video generation costs will increase accordingly.

Sora will appear in ChatGPT

Google has updated its AI for working with documents and Drive.

Gemini now generates text, automatically populates tables with cloud data, creates presentations from scratch, and answers questions about Google Drive content.

It currently only works in English and is only available to Pro and Ultra subscribers. AI-powered file review in Drive is only available in the US.

New ways to create faster with Gemini in Docs, Sheets, Slides and Drive

Advanced Machine Intelligence is developing commercial AI capable of reasoning and planning in the real world. According to LeCun, current LLMs lack this capability.

Lecun founded Advanced Machine Intelligence in 2025. Prior to that, he headed the AI Fundamental Research Lab at Meta since 2013.

Advanced Machine Intelligence was recently valued at $3.5 billion in its funding round.

Yann LeCun has raised $1 billion for his new AI startup

The lawsuit stems from the untrusted supplier status assigned to Anthropic by the US Department of Defense. This status requires companies working with the Pentagon to certify that they do not work with Anthropic.

Anthropic claims the Department of Defense is retaliating against the company for its demands that Claude not be used for mass surveillance or the development of autonomous weapons.

Anthropic is suing the Pentagon

Altman promised to launch erotic chats with ChatGPT at the end of last year. But then OpenAI issued a "code red" due to the release of Gemini 3 and redirected developers to improving ChatGPT.

Now, the "Adult Mode" has been postponed again indefinitely. The reason is roughly the same as last time: to focus on improvements that are a priority for the largest number of users.