powstik
Prove
Growth Paths / AI Research 4: PPO or GRPO on a toy task
ExpertSkills

AI Research 4: PPO or GRPO on a toy task

Implement a policy-optimisation method yourself and test it properly.

Implement a policy-optimisation method yourself and test it properly. Free resources, your code in a public repo, and real work confirmed by people. Part of the Varsity Curriculum: AI Research.

Part of: AI Research · Proves for: Engineers, PG

1 required outcomes6 weeksCredential on completion

Path outcomes

How this is verified: 4 are confirmed by people who saw the work.

SkillsOptional

Learn Reinforcement Learning (AI Engineering Phase 9)

Prerequisite (optional here): the AI Engineering phase this builds on.

Take this Challenge →

2 milestones

  1. 1Do the phase's lessons in public
  2. 2Teach an agent a task
Skills

Implement PPO or GRPO on a toy task

Research step 4.

Take this Challenge →

2 milestones

  1. 1Implement PPO or GRPO yourself
  2. 2Test one design choice

Resources for this path

No affiliate links. No sponsored placements.

Free introduction to policy-gradient methods, including PPO.

Free single-file reference implementations to compare your results against.

Free official docs for GRPO on language models.