EVOKE: Eliciting World Knowledge in Agents for Transferable Decision-Making

Hugging Face Daily Papers Papers
llm-agents post-training decision-making transfer-learning world-knowledge reinforcement-learning

Summary

EVOKE is a post-training method that elicits world knowledge already present in pretrained LLM agents by contrasting alternative goals at fixed states, training policies to rank candidate actions by their consequences rather than contextual habits. It achieves state-of-the-art results across ALFWorld, WebShop, and search-based QA, and transfers well to unseen environments without needing world models or inference-time planning.

Large language models (LLMs) are increasingly deployed as agents for multi-step decision-making, yet transfer poorly to unseen environments. World-model methods address this by training agents to predict future observations, at the cost of additional training and errors that compound when predictions are used for planning. However, for LLM agents operating in digital environments, much of this world knowledge is already internalized during pretraining, which shifts the problem from acquiring it to eliciting it. We argue that typical post-training provides little pressure for such elicitation, since supervision under a single goal at each visited state inadvertently drives policies to rely on superficial contextual habits. We introduce EVOKE, a post-training method that supplies this pressure through goal diversity at fixed states. Motivated by theory showing that an agent competent across diverse goals must encode a world model recoverable from its action preferences, EVOKE holds the environment state and interaction history fixed and ranks the same candidate actions under alternative goals, forcing action preferences to change, so that a policy relying on contextual habits or single-goal correlations cannot order them correctly. This implicitly elicits the policy's pretrained world knowledge to inform decisions. We evaluate EVOKE across diverse tasks in three backbones, demonstrating improved task performance, unseen environment generalization, and data efficiency. We further conduct controlled analyses to better understand what drives these gains. These findings offer a new perspective on eliciting internalized world knowledge for transferable action through direct decision supervision.
Original Article
View Cached Full Text

Cached at: 10/01/26, 08:23 AM

Paper page - EVOKE: Eliciting World Knowledge in Agents for Transferable Decision-Making

Source: https://huggingface.co/papers/2609.38334

EVOKEis a post-training method that elicits the world knowledge already inside pretrained LLM agents, so that they decide by theconsequences of their actionsrather than by contextual habits. It holds the state fixed, swaps in alternative goals, and trains the policy to rank thesame candidate actionsunder each goal, with no world-model module and no inference-time planning.

EVOKE training loop The EVOKE training loop. At collected states, actions are executed and assessed under alternative goals with the state, history, and available actions fixed. The policy learns by contrastive ranking on aggregated preferences; actions can switch between positive and competing across goals.

https://huggingface.co/papers/2609.38334#%E2%9C%A8-highlights✨ Highlights

  • πŸš€Best across the boardβ€” best average on ALFWorld, WebShop, and search-based QA with all three backbones.
  • 🌍Transfers to unseen environmentsβ€”**91.1% / 93.1% / 84.6%**on unseen ALFWorld games (Qwen2.5-3B / 7B / Qwen3-1.7B).
  • 🧠Elicits, not adds, knowledgeβ€” the backbone already encodes action consequences; EVOKE makes the policy decide by the goal rather than by habit.

https://huggingface.co/papers/2609.38334#%F0%9F%92%A1-from-prediction-to-preferenceπŸ’‘ From Prediction to Preference

From prediction to preferenceDo agents need topredictconsequences to use them?

**(a) Predict consequences.**World models learn to predict future observations, adding cost and compounding errors in planning.

**(b) Single-goal supervision.**The knowledge is already in pretrained LLMs, but one goal per state lets the policy fit habits that fail to transfer.

**(c) EVOKE.**Change the goal at a fixed state. Preferences flip, so habits fail and the policy must use its knowledge of consequences.

https://huggingface.co/papers/2609.38334#%F0%9F%93%8A-resultsπŸ“Š Results

https://huggingface.co/papers/2609.38334#main-resultsMain Results

Main results on ALFWorld, WebShop, and search-based QA Main results on ALFWorld and WebShop (left) and search-based QA (right) across three backbones.

ALFWorld unseen results Generalization to unseen ALFWorld games.

https://huggingface.co/papers/2609.38334#analysisAnalysis

All analyses use Qwen2.5-3B on ALFWorld unless noted.

Decision-supervision ablations ***Decision-supervision ablations.*Removing alternative goals or replacing ranking with imitation hurts unseen success. Micro success (%) and average unseen steps, averaged over training seeds.

From knowledge to goal-directed decisions From knowledge to goal-directed decisions.(a) Action consequences are linearly decodable before any ALFWorld training. (b) EVOKE makes the fewest habitual errors, i.e., choosing the action that is correct for another goal in the same context.

Execution on unseen games ***Execution on unseen games.*EVOKE succeeds earlier and wastes fewer actions: (a) cumulative success over steps, (b) invalid actions per game, (c) revisits per successful game.

https://huggingface.co/papers/2609.38334#%F0%9F%93%9D-citationπŸ“ Citation

@article{guo2026evoke,
  title={EVOKE: Eliciting World Knowledge in Agents for Transferable Decision-Making},
  author={Guo, Yuhan and Liu, Jinming and Xu, Liang and Li, Ziqiang and Huang, Jianguo and Wang, Zhicheng and Zhu, Hu and Chen, Qiuyu and Wei, Yuntao and Jin, Xin and Zeng, Wenjun},
  journal={arXiv preprint arXiv:2609.38334},
  year={2026}
}

Similar Articles

Alaya-EVOKE: From Linear-Scaling Supervision to Endless World

Hugging Face Daily Papers

Evoke is an interactive world model with external persistent memory and a redesigned long-horizon teacher, enabling open-ended video generation with bounded context and low latency. It achieves state-of-the-art performance on WBench while staying competitive on VBench-Long and VBench-2.0.

Rethinking Experience Utilization in Self-Evolving Language Model Agents

arXiv cs.CL

This paper introduces ExpWeaver, a framework that optimizes how self-evolving language model agents utilize past experiences during runtime decision-making. It demonstrates that selectively invoking experience based on reasoning uncertainty improves performance across various environments and models.

Agent-Editing World Model: Rethinking World Modeling for LLM Agents

Hugging Face Daily Papers

This paper proposes the Agent-Editing World Model (AEWM) to enhance LLM agents by modeling how reasoning and actions shape future task progress, achieving significant performance improvements on benchmarks across domains like search and software engineering.