News of the day
1. Nvidia releases Nemotron 3 AI models (Nano, Super, Ultra) with open-source data, MoE architecture for speed, and a 1M token context window. → Read more
2. Google Translate beta now uses Gemini for real-time voice translation, preserving original tone and rhythm for seamless communication. → Read more
3. MIT Technology Review launches 'AI Hype Correction' package, critically examining AI's true capabilities and future impact beyond inflated claims. → Read more
4. Vibe Coding Luxembourg successfully launched 7 live projects in under two days with 7 participants of diverse backgrounds. Next session Feb 9-10, 2026. → Read more
Our take
Hi Dotikers!
Nvidia is no longer content just selling shovels to gold miners. The GPU giant just unveiled Nemotron 3, its new open source AI model family available in three sizes: Nano, Super, and Ultra. An announcement that marks a rather clever strategic shift.
What makes this release interesting isn't just the raw performance, though it delivers. The Nano model boasts 4x higher throughput than its predecessor and cuts reasoning tokens by 60% thanks to a hybrid architecture combining Mamba, Transformer, and Mixture of Experts. A one million token context window rounds out the package, allowing developers to process entire codebases or massive documents without chunking.
But the real masterstroke lies elsewhere. Nvidia is publishing everything: three trillion tokens of pre-training data, 18 million post-training samples, and even reinforcement learning training environments through NeMo Gym. This level of openness is unprecedented in the industry.
The early adopter list speaks for itself: Cursor, Perplexity, ServiceNow, Palantir. The players betting seriously on agentic AI are already on board.
Nvidia is methodically building an ecosystem where it controls the hardware, provides the models, and now the training data. While everyone else is playing chess, Jensen Huang is clearly playing an entirely different game. Only the Nano model is available now, with Super and Ultra expected in the first half of 2026.
G.
