PPO → DPO → GRPO→ Rubrics
AI by Hand ✍️ Seminar Series
In last week’s AI by Hand ✍️ seminar, I talked about reinforcement learning from first principles. I started all the way back from pre-training and inference, before climbing forward through the major evolution stages of the modern RL stack: PPO → DPO → GRPO. I finally arrived at Rubrics, widely considered one of today’s frontier topics in RL research.
I was joined by Cameron R. Wolfe, author of the Deep (Learning) Focus newsletter (60K subscribers) and a Senior Research Scientist at Netflix. He gave a special guest lecture based on his popular recent article Rubric-based Rewards for RL.
After eight seminars in 2026, I’ve settled into a format that has clearly resonated:
AI by Hand Lecture → Industry Guest Interview → Industry Guest Lecture.
I’m the professor in the first half—and the student in the second. 🙌
Feedback
Industry Expert: Cameron Wolfe
As a professor in academia, how can I possibly know what kind of research is really going on in the industry? I turn to people like Cameron. I've been reading his long-form articles long before I even started sharing my own content through this AI by Hand newsletter. Cameron was one of the early supporters who encouraged me to share more publicly.
I still remember two years ago when he told me he had just gotten married and was moving to Netflix to become a Senior Research Scientist working on reinforcement learning. He mentioned he might need to pause the Deep Focus newsletter for a while. Fortunately for all of us, the break was short. Cameron kept writing. I kept learning.
In the seminar, I finally got a chance to interview him so his story can be heard by all of you:
Q: Can you use Jiu-Jitsu as a metaphor to explain reinforcement learning?
Q: Reinforcement learning used to feel like robotics demos. Why has RL become so important for frontier language models?
Q: What are the main challenges of doing RL research at large scale?
Q: What motivated you to write the Deep Focus newsletter?
Watch the recording to hear Cameron explain it in his own words.
Recording
00:00 Welcome & Introduction
03:53 Pretraining
07:32 Inference
10:56 Proximal Policy Optimization (PPO)
23:58 Direct Policy Optimization (DPO)
28:12 Group Relative Policy Optimization (GRPO)
31:55 Guest Interview: Cameron Wolfe
39:42 Guest Lecture: Rubrics for RL
Excel Workbook
Previous Seminars
OpenClaw - 12 Stages of Evolution from the Transformer (2/19/2026)
Transformer - Six Levels of Understanding (2/12/2026)
Meta Superintelligence Labs vs Facebook AI Research (2/5/2026)
9 AI Eval Formulas (1/29/2026)
Google Ironwood TPU: From Bits to HBM (1/22/2026)
How AWS Uses Small Models Learn Tool Use (1/15/2026)
Attention (1/15/2026)
DeepSeek’s Manifold-Constrained Hyper Connection (mHC) (1/8/2026)
Introduction to Generative AI (1/8/2026)
Gated Attention (NeurIPS 2025 Best Paper) (12/16/2025)
Could you please share the link to exact last week's material ( reinforcement learning from first principles) that you are mentioning in this post ?