CLPO: Curriculum Learning meets Policy Optimization for LLM Reasoning

Kari Jaaskelainen

30 Sep 2025 — 1 min read

A dynamic, self‑paced curriculum that restructures problems to match model ability in RLVR

Online RL with Verifiable Rewards (RLVR) has boosted LLM reasoning—but most methods treat all problems equally, wasting effort on solved items and flailing on those beyond current capability. CLPO fixes that with a dynamic pedagogy: it continuously assesses difficulty from rollouts, then restructures the training set so the model learns at its frontier.

Two pillars drive CLPO. First, Online Curriculum: real‑time difficulty estimation steers sampling toward the most informative items. Second, Adaptive Problem Restructuring: the model acts as its own teacher—diversifying medium items to promote generalization while simplifying hard ones into attainable steps. This transforms static optimization into a co‑evolving loop where data and policy advance together.

Across eight challenging math and general reasoning benchmarks, CLPO achieves state‑of‑the‑art results, delivering an average pass@1 improvement of 6.96% over competitive baselines. The gains suggest that pacing and structure—core tenets of human pedagogy—translate into more efficient exploration and higher ceilings for LLM reasoning under verifiable signals.

Why it matters: curriculum isn’t just a data trick—it’s a control layer over learning dynamics. For practitioners, CLPO offers a principled way to concentrate compute where it counts and unlock harder problems sooner.

Paper: arXiv: CLPO
Register: AiFeta

#RLVR #CurriculumLearning #LLM #Reasoning #PolicyOptimization #MathAI #AITraining

From f(x) and g(x) to f(g(x)): LLMs Learn New Skills in RL by Composing Old Ones

Evidence that RL teaches genuinely new abilities: compositional skills emerge and transfer across tasks Does RL merely reweight what an LLM already knows—or can it teach genuinely new skills? This paper offers concrete evidence for the latter. Using a controlled, synthetic framework, the authors define “skills” as string transformation

OpenGPT-4o-Image: A Comprehensive Dataset for Advanced Image Generation and Editing

A structured 80k instruction–image corpus spanning 11 domains and 51 subtasks to train unified visual editors Unified models for image generation and editing hit a data ceiling: existing corpora emphasize basic manipulations but miss real‑world complexity. OpenGPT‑4o‑Image tackles this with a hierarchical task taxonomy and automated

Random Policy Valuation is Enough for LLM Reasoning with Verifiable Rewards

ROVER replaces PPO loops with uniform‑policy Q‑values—boosting quality and diversity in math reasoning Popular RLVR methods for LLM reasoning lean on generalized policy iteration (e.g., PPO/GRPO), but suffer instability and diversity collapse. This paper reframes math RLVR as a specialized finite‑horizon MDP with deterministic

Scaling Synthetic Task Generation for Agents via Exploration

AutoPlay explores environments to synthesize diverse, verifiable tasks—fueling stronger UI agents Agent post‑training hinges on rich, grounded tasks—but human‑curated datasets are costly and often shallow. AutoPlay introduces a scalable alternative: let an MLLM‑based explorer systematically probe interactive environments, discover states and affordances, and then synthesize

Read more

From f(x) and g(x) to f(g(x)): LLMs Learn New Skills in RL by Composing Old Ones

OpenGPT-4o-Image: A Comprehensive Dataset for Advanced Image Generation and Editing

Random Policy Valuation is Enough for LLM Reasoning with Verifiable Rewards

Scaling Synthetic Task Generation for Agents via Exploration