DeepSeek R1 Incentivizing Reasoning Capability in LLMs via Reinforcement Learning (paper explained)

[@PoleesuSrinivasCh]

A Better explanation even which will lead in all Sources.

[@SapienSpace]

Fascinating review. I glanced at the paper, particularly at GRPO and GAE. GRPO looks a lot like Fuzzy-Logic with nodes or attention heads adapted to experience (e.g. such as "relative" via using K-means group clustering).
Looking more deeply at GAE (Generalized Advantage Estimation) it is for an adaptive control system.
I would not be surprised if the origin of the deep learning usage of Theta is an angle of a pendulum.

[@SapienSpace]

Overlapping membership functions used in Fuzzy Logic is very similar to KL.

[@AIBites]

Don't have much experience with fuzzy logic. But I like your perspective 🙂

[@KhurramXahiL-py5dq]

Great explanation

[@AIBites]

Thanks 👍

[@francesclopez6192]

Thank you for your explanation !

[@AIBites]

My pleasure 😊

[@mukeshreddy7909]

great video

[@AIBites]

Thanks!

[@amortalbeing]

thanks a lot.

[@AIBites]

Most welcome!