RL Verifiers by Yash More (Cerebras)

​Yash will give an overview of RLVR and verifiers.

​Reinforcement Learning with Verifiable Rewards (RLVR) replaces noisy human feedback with deterministic signals that make verification robust. In this talk, we will examine what makes a reward verifiable, how credit can be effectively assigned through process and outcome supervision, and the algorithms that enable RLVR to scale and shape reasoning capabilities of LLMs.

​Some materials (for better context):
- ⁠DeepSeekMath (https://arxiv.org/abs/2402.03300)
- Tulu3 (https://arxiv.org/pdf/2411.15124)
- https://arxiv.org/pdf/2408.03314

​There will be snacks and drinks.