AI Paper Atlas

Research map / Multi-agent systems and games / Human-AI and goal-conditioned

A Behavioural and Representational Evaluation of Goal-Directedness in Language Model Agents

Raghu Arghal, Phoebe Chen, Niall Dalton, Evgenii Kortukov, Calum McNamara, Angelos Nalmpantis, Moksh Nirvaan, Gabriele Sarti, Mario Giulianelli

ICML 2026 · Multi-agent systems and games › Human-AI and goal-conditioned · approach: Language-grounded reinforcement agents · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (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 changedWhereMeasuredEffect
Increase grid size and obstacle densityGPT-OSS-20B agent · MiniGrid navigationper-action accuracyHurt
Apply iso-difficulty transforms (reflect, rotate, swap start/goal, transpose)GPT-OSS-20B agent · MiniGrid navigationbehavioural metricsNo clear effect
Add a functionless key (door removed)GPT-OSS-20B agent · KeyNoDoorEnvper-action accuracy (%): 98.7 -> 97.2Hurt
Add vestigial key on one of two optimal pathsGPT-OSS-20B agent · 2PathKeyEnvsuccess rate (%): 75.5 -> 71.4Hurt
Add vestigial key on one of two optimal pathsGPT-OSS-20B agent · 2PathKeyEnvper-action accuracy (%): 74.3 -> 76.0Improved
Use MLP probe instead of linear probe for cognitive maps (ablation)probe on layer-15 pre-reasoning activations · 11x11 gridscell-type accuracy (%): 39.1 -> 75.7Improved
Decode after reasoning instead of before (ablation)cognitive map MLP probe · grid cognitive mapoverall probe accuracy (%): 75 -> 60Hurt
Decode plans from post-reasoning instead of pre-reasoning activations (N=1) (ablation)2-layer plan decoder · plan decodingprefix accuracy (%): 41.5 -> 66.49Improved
Decode plans from pre-reasoning instead of post-reasoning activations for long prefixes (ablation)2-layer plan decoder · plan decoding, N>=5prefix accuracyImproved
Increase plan decoder depth from 2 to 4 layers (ablation)plan decoder · plan decodingprefix accuracyNo clear effect
Evaluate actions against top-k decoded grids instead of top-1 (ablation)decoded cognitive maps · action optimalityaction accuracyImproved
Patch activations at a single layer on chat template tokens (ablation)GPT-OSS-20B · activation patchingaction distributionNo clear effect

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy