All videos
0:00 / 0:00
research

GLM-5.2: DeepSeek Was Wrong About RL?

bycloud23 July 2026Watch on YouTube

What you'll learn

  • Zhipu AI switched back from GRPO to PPO for reinforcement learning training, despite previous preference for GRPO
  • The GLM-5.2 technical blog contains insights about training methodology that go beyond model performance alone
  • PPO proved more effective in practice for training the GLM-5.2 model than the previously used GRPO approach

Frequently asked questions

Why did Zhipu AI switch back from GRPO to PPO?
Although GRPO was used previously, PPO proved more effective in practice for training GLM-5.2. Zhipu AI's technical blog covers this decision and shows that PPO works better for their model training.
What makes the GLM-5.2 blog interesting according to the video?
The blog contains deep insights about training methodology that many people don't discuss. It goes beyond just model performance and covers important details of how the model was trained.
Which reinforcement learning methods are compared in GLM-5.2?
The video analyzes the choice between GRPO and PPO as reinforcement learning approaches for the GLM-5.2 model. The eventual selection of PPO and the reasons behind it are central to the discussion.

Topics

large language modelsreinforcement learningtraining methodsglm-5.2model architecture

Read next

Description from the channel

Need high quality cloud GPUs? Check out Verda now, and use the code BYCLOUD-50 to get $50 of compute credits for just $5! https://verda.com/?utm_source=bycloud&utm_medium=referral&utm_campaign=sponsorship&utm_content=deepseekv4 Late to the GLM-5.2 party, but I think there is actually much more insights in their blogs I don't see people talk about than the model performance itself. Especially the part where they decided to stop using GRPO and go back to PPO! my latest project: Intuitive AI Academy We just wrote a new piece on Optimization!! https://intuitiveai.academy/ limited time code "LOCKIN" for 35% off yearly plan My Newsletter (weekly top research papers) https://mail.bycloud.ai/ My Patreon https://www.patreon.com/c/bycloud GLM-5.2 [Blog] https://z.ai/blog/glm-5.2 Try out my new fav place to learn how to code https://scrimba.com/?via=bycloudAI This video is supported by the kind Patrons & YouTube Members: 🙏Spam Maj, Alex, Chris LeDoux, DX Research Group, Poof N' Inu, Deagan, Robert Zawiasa, Ryszard Warzocha, Tobe2d, Louis Muk, Akkusativ, Kevin Tai, Mark Buckler, NO U, Tony Jimenez, Ângelo Fonseca, jiye, Anushka, Asad Dhamani, Binnie Yiu, Calvin Yan, Clayton Ford, Diego Silva, Etrotta, Gonzalo Fidalgo, Handenon, Hector, Jake Disco very, Michael Brenner, Nilly K, OlegWock, Daddy Wen, Shuhong Chen, Sid_Cipher, Stefan Lorenz, Sup, tantan assawade, Thipok Tham, Thomas Di Martino, Thomas Lin, Richárd Nagyfi, Paperboy, mika, Leo, Berhane-Meskel, Kadhai Pesalam, mayssam, Bill Mangrum, nyaa, Toru Mon, Lame Plane, Matej Macak, Len Mo, saylikhapekar, ZyanSheep, THEVIERAOS, Ricardo Raphael Corona-Moreno, superchordate [Discord] https://discord.gg/NhJZGtH [Twitter] https://twitter.com/bycloudai [Patreon] https://www.patreon.com/bycloud [Business Inquiries] bycloud@smoothmedia.co [Other Inquiries] bycloudai@gmail.com [Profile & Banner Art] https://twitter.com/pygm7 Manim Animations created by https://x.com/miyashita_03 [Ko-fi] https://ko-fi.com/bycloudai