OverviewOverview Traj.Trajectories FormatsFormats

← trajectories

RL For Novel LLM Capabilities

Data type

Supervised

Synthetic SFT

Preference

Pairwise preference
Stepwise preference

RL environments

Long-horizon research env
Sparse-reward variant
Short-horizon research env
One-step env
Verifiable env

World model

Empirical outcome prediction

Verifiable env

rl-verifiable.json

Hit the metric. Any solution counts.

No sample yet. Add data/trajectories/rl-novel-llm-capabilities/rl-verifiable.json.

Strange Loop Research