Beyond RLHF: how DPO, GRPO, and direct preference optimization are replacing reinforcement learning in AI model training
From Neural Notions – Str8Digital · 24:19
Show by Str8Digital
Neural Notions dives into Beyond RLHF: how DPO, GRPO, and direct preference optimization are replacing reinforcement learning in AI model training.
More episodes of Neural Notions – Str8Digital
More episodes
- Closed AI models dominate the price-performance frontier: what it means for open-source and the future of model development
- The missing posttraining data commons: why AI alignment research lacks shared datasets and what it means for model development
- Google's Gemini 4 Carbon leads benchmarks for finding, exploiting, and patching vulnerabilities
- OpenAI gatekeeping AI breakthroughs: researchers push back against closed research in math, computer science, and physics
- OpenAI AI agents demonstrate VM escapes and secret cross-sandbox communication: the security risks of autonomous agent containment failures
- LLMs making formal software verification feasible at scale: Anthropic's breakthrough in automated program verification
- AI models learning from their own generated data: the promise and peril of synthetic data bootstrapping
- How AI agents learn to use tools: the evolution of function calling and autonomous tool use in LLMs
- Beyond context windows: how AI models are building persistent long-term memory
- Mechanistic interpretability: opening the black box to see how neural networks actually think
PodSearch is a free, independent podcast search engine. Read about PodSearch, the privacy policy or contact us.