Note de episod
What if adding math data actually improves reasoning in non-math tasks?
In this episode of #WiAIRpodcast, Syeda from the Carnegie Mellon University and NVIDIA presents Nemotron CrossThink (EACL 2026 Oral) - a framework that challenges how we train LLMs for reasoning.
Paper: https://arxiv.org/abs/2504.13941
Key insights you don’t want to miss:
• Why multi-domain training beats specialized models
• The surprising role of math in improving general reasoning
• How to get +performance with ~40% less data
• Why harder training samples are better than more data
• How LLMs learn adaptive…
Site: podcasters.spotify.com

