Research map / Large language models / Agents, code and math
Hierarchical Decision Making by Generating and Following Natural Language Instructions
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use latent natural-language instructions and a separate executor instead of directly imitating actions | MiniRTS agent · MiniRTS | win rate | Helped |
| Encode instructions compositionally with an RNN rather than as one-hot instruction IDs | Executor · MiniRTS | negative log likelihood: 3.05 $\pm$ 0.0015 -> 2.88 $\pm$ 0.0006 | Helped |
| Encode instructions compositionally with an RNN rather than as one-hot instruction IDs | Executor · MiniRTS | win rate: 49.6 -> 57.9 | Helped |
| Use a bag-of-words instruction encoder instead of one-hot instruction IDs | Executor · MiniRTS | win rate: 49.6 -> 54.2 | Helped |
| Use the RNN discriminative instructor rather than the one-hot instructor | Instructor (N=500 instructions) · MiniRTS | win rate: 43.1 -> 57.9 | Helped |
| Use a larger instruction set, from N=50 to N=500 | RNN discriminative instructor · MiniRTS | win rate: 47.8 -> 57.9 | Helped |
| Use a larger instruction set, from N=50 to N=500 | One-hot instructor · MiniRTS | win rate: 44.6 -> 43.1 | Mixed |
| Use a generative RNN instructor rather than a discriminative RNN instructor | Instructor (N=500 instructions) · MiniRTS | win rate: 57.9 -> 54.8 | Hurt |
| Attend to the most recent 5 instructions rather than use only the current instruction (ablation) | RNN executor · MiniRTS | win rate: 52.9 -> 57.9 | Helped |
| Attend to the most recent 5 instructions rather than use only the current instruction (ablation) | RNN executor · MiniRTS | negative log likelihood: 2.93 -> 2.88 | Helped |
Closest papers by meaning
- Emergent Hierarchical Reasoning in LLMs through Reinforcement Learning (ICLR 2026 · similarity 0.88)
- Enhancing Decision-Making of Large Language Models via Actor-Critic (ICML 2025 · similarity 0.87)
- Ground-Compose-Reinforce: Grounding Language in Agentic Behaviours using Limited Data (NeurIPS 2025 · similarity 0.87)
- Language Control Diffusion: Efficiently Scaling through Space, Time, and Tasks (ICLR 2024 · similarity 0.87)
- $\texttt{Multi}^2$: Hierarchical Multi-Agent Decision-Making with LLM-Based Agents in Interactive Environments (ICML 2026 · similarity 0.87)
- From Words to Actions: Unveiling the Theoretical Underpinnings of LLM-Driven Autonomous Systems (ICML 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.