We just made LLM inference up to 30% faster ⚡
Introducing DNeo, a new diffusion drafter designed in collaboration with @cerebras researchers.
The secret? Your draft model needs a working memory, and the "right" context encoding mechanism.
📄 Paper: https://t.co/Pj1qpeqpO9
For 33 years, Jensen Huang told his employees the company was thirty days from going out of business.
It never happened. He braced for every ending except the one he got.
At nine, his parents sent him to America. The boarding school an uncle found in rural Kentucky turned out to be a school for troubled kids. He was the youngest one there, and he was given the bathrooms to clean.
At fifteen he was at Denny's. Dishwasher, then busboy, then waiter.
At thirty he sat in a Denny's booth with two friends and sketched a chip company. Within a few years it was nearly out of money, and he laid off about half the staff to keep it alive.
The thirty days line started there. He kept saying it long after it stopped being true, because the fear never left him.
The fear made him early. He launched CUDA in 2006 and funded it through years of doubt. In 2012, the deep learning era began on two of his graphics cards.
He guarded the company against everything. Rivals. Crashes. Complacency.
The thing that ended it was never on the list.
On November 8, he wrote to every employee that he had been unfaithful to his wife. "There is no excuse for what I did."
The spy allegations around it are unproven. The affair is not.
Four days later, he resigned.
On November 17, he said he would be honored to return. That afternoon, Nvidia reported the greatest quarter in its history. $115.8 billion in revenue. $65.3 billion in profit. The chairman opened the call by saying the board does not intend to bring him back.
More than 200,000 people signed a letter asking for him. At the first company meeting without him, the most common question from employees was about pay.
Nobody betrayed him. The machine did exactly what he built it to do. He spent three decades making sure the company could survive anything.
Anything turned out to include him.
Every founder builds two things. The company the world audits every quarter, and the life underneath it that nobody audits at all.
He built the first one better than anyone alive.
His statement ended with a line written for no board and no shareholder. He told his family, "I will spend the rest of my life being worthy of yours."
For 33 years, he lived thirty days at a time.
That promise has no deadline. No quarter will measure it, and nobody will ever report whether he kept it.
Nvidia will find another CEO. He built it so it could.
His family does not get another him.
Envato is making MiniMax H3 free and unlimited for 24 hours.
It takes text, images, video, and audio together to generate up to 15s clips at 2K with native stereo sound, built for strong instruction following and commercial-style output.
Temporary open access to a capable multimodal video model lets more people test high-quality generation in real workflows.
Must-read papers of the week
▪️ World In-Context Learning for Agentic Robot Manipulation
▪️ Don’t Just Trust Your Memory
▪️ Is Your Robot’s Memory Just a Clock? Temporal Shortcuts in Memory-Augmented VLA Policies
▪️ Unbundling Multimodal Vision Encoding
▪️ V-JEPA-LM
▪️ Learning Heterogeneous Tactile Representation for General Dexterous Manipulation
▪️ Training Reasoning Models in Simulation to Natively Use Real Robots
▪️ Embodied First Exam
▪️ Continual-ARC: Measuring Skill Acquisition over a Working Lifetime
▪️ Reframe: Observation-Aligned Multi-Teacher On-Policy Distillation for Agentic Visual Reasoning
▪️ MemInStream: Memorizing or Merely Recognizing? Rethinking Streaming Video Memory Evaluation
▪️ Infrathon: Benchmarking Language Agents for Real-World, Long-Horizon AI Infra Optimization
▪️ Actor, Judge, or Teacher? Understanding Decision Models in Long-Horizon Agents
▪️ GTAGym: AAA Game Tasks Environment for Evaluation and Reinforcement Learning
Explore these to keep up with main AI trends. Here’s also the full list of outstanding papers + links, and our weekly AI news digest: https://t.co/XJLza8hzI8
things i wish someone told me earlier:
> cgpa matters, but nobody cares if you have a 9.5 and zero actual skills.
> build random stuff. most of what i learned came from projects, not lectures.
> join clubs, meet people, make connections. don't just collect certificates.
> apply for internships even if you think you're underqualified. let them reject you.
> talk to seniors. some of the best advice i've received came from random conversations.
> stop treating linkedin like a competition. everyone's figuring things out.
> don't spend 4 years preparing for a life you forget to actually live.