Artificial Analysis 1/4 We see no wall in post-training. Scaling RL software, infra, and data keeps yielding major capability gains. We trained across 30 RL environments with up to 4,000 instances per batch — math, code, STEM, agentic tool use, SWE, terminal, safety — all in a unified

RL is exploding right now - and for good reason, RL stages unlock capability gains in models that power agents that “work” work. If you’re at #NVIDIA #GTC26, come see Chris 🇨🇦 and me live at our Connect with the Experts session. We’ll talk about this and more - - Plug RL

Think long term. Using a frontier model for every task is adorable but the real flex is rocking a team of specialists - improve cost while retaining accuracy.

At #NVIDIAGTC 2026, learn how to build domain-specific, tool-using AI agents 🤖 using NVIDIA NeMo Gym and NeMo RL open libraries with RLVR on NVIDIA Nemotron! Wednesday, March 18 | 1:00 p.m. PT Explore my session: nvda.ws/4qHZia0

Sandwich falls apart mid-bite. Tech Twitter: “MCP is dead”

Some of the biggest “intelligence” gains in frontier models are coming from scaling RL across diverse environments in post-training. Nemotron 3 Super is a strong example: 21 environment configurations, and ~1.2M rollouts covering a broad set of agentic tasks. These workflows are

Delighted to see that you find it useful, Akshay 🚀. We put a lot of love and effort into cooking it!

When you only have 20 minutes of material for a 1 hour presentation 🤣 (looking for a gift for my 1 yo nephew)

He gets it. Obviously.

Double down on Team Green. 💚☘️ NVIDIA #GTC26 next week, but also St. Patrick’s. Why not GPUs by day and Guinness by night. 🍻

All ready for the big show! #NVIDIA #GTC A Jam packed SAP center

Less than 30 minutes left in the world’s shortest hackathon ⏱️ at #NVIDIA #GTC2026 2 hrs to vibe code an agentic AI app, that uses Nemotron. Honored to be judging alongside Two Minute Papers Dr. Karoly Zsolnai-Fehér and an incredible panel of judges.

Joe Nemotron demonstrates exactly: 1. How to attract an audience for some of the best sessions this GTC. 2. What to do when encountering a mountain lion. That energy is ⚡️

You better believe it

Engineers at top companies used to flex free lunches and nap pods - soon they’ll be bragging about their ‘token luxury’. The real flex is unlimited tokens as the ultimate force multiplier.

Chaos isn’t as bad as we think

North Cal coast: impossible to describe. Mendocino: still underrated. Me: yearly pilgrimage, watching grey whales. 🐋

I trained this LTX LTX 2.3 LoRA of George Costanza at home on my 5090 in about a day with AI Toolkit. I generated this 30 second video with ComfyUI on my 5090 in 6 minutes. Open source is, always has been, and always will be, the future of generative AI. (SOUND ON)