Why does an NVIDIA engineer insist open weights change everything for developers?

Loud advocate for openness

Alexiuk runs a YouTube channel where thousands of developers learn to build, fine-tune, evaluate and ship LLM applications and agents. "The people who will most help us understand how to build and apply AI likely haven't even heard of it yet," he says. His rule: if you believe in open source AI, be loud about it. On LinkedIn he calls himself "No. 1 Nemotron Fan." On X he uses the handle llm_wizard with a cartoon wizard avatar. This directness makes his teaching style appealing.

What the Nemotron family is

Nemotron is a line of open multimodal models built for long-running agents that self-evolve, as well as specialized AI agents. Alexiuk compares it to a magical library: "You're a wizard in search of prepared spells, and Nemotron has a spell for everything." The lineup includes Nemotron 3.5 Lightning for repetitive high-frequency execution work. Nemotron Ultra targets autonomous agents; it also handles deep research and enterprise tasks. He argues transparency isn't a feature of good AI — it's the precondition. "Fully open source AI is necessary for the democratization of intelligence," he adds. You should have access to the weights, the data, the training recipes, everything.

From physics to a bird classifier

The fascination with machine learning started in a computational physics class at Trent University in Ontario. Alexiuk learned about TensorFlow and that you could build image classifiers running on a GPU. His first pet project: a classifier that alerted him whenever his friend's Yellow-Faced Gray Budgerigar appeared on webcam during shared gaming sessions. "I wasn't focused on my friend's webcam, but I didn't want to miss seeing the bird," he recalls. That moment changed everything. He realized what technology could do.

Where to watch

The NVIDIA Nemotron Labs Livestreams series on YouTube features Alexiuk regularly. Jensen Huang's GTC Berlin keynote recording is also available from October 21, 11 a.m.–1 p.m. CEST. The announcement does not say how many developers already use Nemotron in production.