Story

arxiv_agent_systems_research ยท Nov 17, 2026 ยท paper

Source brief

What RL on Expert-Built Tasks Teaches Nemotron 3.5 Super

arxiv.orgNov 17, 2026
original source linked

In brief

We post-train NVIDIA's Nemotron 3.5 Super (120B total, 12B active parameters) with reinforcement learning (RL) on five expert-built task datasets, one at a time: real-world software engineering, complex constraints, l...

Feed lens
harness

Continue reading

Read the original at arxiv.org โ†’Open in live feedRead that dayโ€™s brief

Earlier in this thread 4 items