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

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

REVIEW 4 major objections 4 minor 8 references

The paper argues that goal-directedness in an LLM agent requires both behavioural evaluation and probing of internal representations, and shows the agent encodes a coarse cognitive map and plans, with reasoning shifting from spatial cues to

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-03 03:04 UTC pith:NRPU7DSD

load-bearing objection Solid behavioral package, real methodological novelty, but the representational claims are not yet secured: the probing section needs basic controls before the cognitive-map story holds. the 4 major comments →

arxiv 2602.08964 v2 pith:NRPU7DSD submitted 2026-02-09 cs.LG cs.AIcs.CLcs.CY

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

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

show 1 more author

Mario Giulianelli

This is my paper

classification cs.LG cs.AIcs.CLcs.CY

keywords goal-directednessLLM agentsinterpretabilityprobingcognitive mapsgrid world navigationplan decodingreasoning

verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper is trying to establish that goal-directedness in an LLM-based agent can be assessed only by combining behavioural evaluation with inspection of the model's internal representations, and it demonstrates this in a grid-navigation case study. The agent's actions are mostly consistent with the optimal policy, robust to difficulty-preserving transformations and instrumental goals, but systematically biased by goal-like but irrelevant objects such as keys. Probing hidden activations shows the agent non-linearly encodes a coarse spatial map of the grid, with approximate positions for itself and the goal, and that these representations reorganise during reasoning: broader spatial and long-horizon plan information decays while the next action becomes more decodable. The sympathetic reader should care because this supplies a concrete method to distinguish failures of goal-directedness from failures of capability and from faulty internal world models.

Core claim

The central claim is that the LLM agent non-linearly encodes a coarse spatial map of the environment, preserving approximate task-relevant cues about its position and the goal location; that its actions are broadly consistent with these internal representations; and that reasoning reorganises them, shifting from broader environment structural cues toward information supporting immediate action selection. Decoding cell identities with an MLP probe reaches about 70% accuracy versus about 39% for a linear probe, showing the encoding is non-linear. Goal and agent positions are localised with high recall but blurred precision, and goal-distance is decodable with a mean absolute error of roughly t

What carries the argument

Two central objects carry the argument. The first is the cognitive-map probe: an MLP classifier that, given a residual-stream activation and the (x,y) coordinates of a queried cell, predicts the cell type (agent, goal, wall, open, padding), allowing reconstruction of the model's decoded belief over the grid state. The second is a one-shot plan decoder: a Transformer decoder that predicts an entire T-step action sequence simultaneously from a fixed set of token activations, without autoregressive conditioning, ensuring that any recovered multi-step structure must already be present in the model's representations rather than manufactured by the probe's own inference. The behavioural component

Load-bearing premise

The cognitive-map and consistency claims assume that probes trained on ground-truth labels with (x,y) coordinate conditioning recover the model's internal beliefs, rather than exploiting shortcuts such as the grid text literally present in the input, coordinate priors, or probe capacity; no control probes or ablations removing the coordinate conditioning are reported.

What would settle it

Train the same probe on activations collected from grids where the goal and agent characters are replaced by neutral symbols (or on random feature vectors with identical coordinate conditioning); if per-class accuracy stays well above chance and localisation precision persists, the internal-map claim collapses. A simpler control: remove the (x,y) coordinate input from the probe and check whether accuracy drops to chance; if it does not, the coordinates are the carrier.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

The agent's performance scales with grid size and obstacle density and is robust to difficulty-preserving transformations, indicating that navigation behaviour is driven by task-relevant structure rather than incidental grid configurations.
The agent reliably solves instrumental key-door subtasks but is systematically attracted to goal-like reward-irrelevant objects, detouring toward a useless key in 75% of non-optimal actions and favouring the key-containing path in 67.3% of two-path trials.
Probes decode a coarse non-linear cognitive map of the environment from hidden activations, with approximate but localisable agent and goal positions, supporting the claim that goal-relevant spatial beliefs are carried in the model's internal state.
Actions are broadly consistent with decoded beliefs: agreement between the ground-truth-optimal policy and the decoded-map-optimal policy averages about 84%, and a substantial share of suboptimal actions are optimal with respect to the agent's own faulty map, attributing many failures to fuzzy world representations rather than to a lack of goal-directedness.
Reasoning re-weights internal representations: pre-reasoning activations support longer-horizon plan decoding (e.g., 13% vs 9% prefix accuracy at N=4), while post-reasoning activations boost next-step decodability (53.9% vs 40.3% at N=1), indicating a shift from environment-structural cues to immediate action selection.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

If the probes are reliable, this framework offers a way to adjudicate capability failure versus goal-directedness failure in real deployments: an agent that acts optimally with respect to its decoded beliefs but poorly relative to the environment is rational given a faulty model, which changes how one should intervene.
The key-attraction bias suggests a general phenomenon: LLM agents may treat semantically salient objects from pretraining as goal-relevant even when the current task specification says otherwise; this could be tested in more naturalistic environments with arbitrary distractor objects.
The pre- and post-reasoning representational shift implies that reasoning partly acts as a compression from world-model to policy features. A testable extension is to vary reasoning effort and measure whether the crossing point in plan decodability moves, predicting that higher reasoning effort compresses spatial cues more aggressively.
The coarse, blurred localisation (high recall, low precision) mirrors place-field-like uncertainty in biological navigation; one could test whether the decoded map's uncertainty correlates with the agent's behavioural entropy or distance-to-goal, connecting the representation to calibration of internal beliefs.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Desk Editor's Note

Solid behavioral package, real methodological novelty, but the representational claims are not yet secured: the probing section needs basic controls before the cognitive-map story holds.

read the letter

You should know two things. First, the behavioral half of this paper is genuinely solid: the iso-difficulty transformations, the multi-goal variants, and the key-artifact bias results are cleanly measured with paired Wilcoxon tests and random grid generation. Second, the representational half—which is the real contribution—has a load-bearing hole: the probes lack the right controls, so the "cognitive map" and the claim that actions are "broadly consistent with internal representations" are not yet established.

What is actually new: the integration of behavioral robustness tests with internal-state probing in one framework, the pre/post-reasoning decodability crossing in the plan decoder, and the observation that goal-like artifacts (a key) attract the agent even when functionally useless. The plan decoder design is thoughtful—one-shot decoding rather than autoregressive avoids a real confound, and the increasing gap from the random baseline is a meaningful observation. The paper also deserves credit for being honest about its limitations, including the partial-observability failures in the appendix.

The soft spots are concentrated in §5.1 and §5.2. The probes are trained on activations taken after the model has seen the full grid text, and they are conditioned on (x, y) coordinates. Without a padding-majority baseline, a coordinate-only baseline, or control probes on shuffled activations or random features, the 70% MLP probe accuracy could be largely a nonlinear readout of the input grid plus coordinate priors rather than evidence of an internal spatial map. The same issue undermines the "Rec." metric in §5.2: attributing suboptimal actions to "fuzzy world representations" is circular if the fuzziness comes from probe error rather than the model's actual beliefs. The plan decoder is less exposed because the one-shot design prevents the probe from generating trajectory structure, but it still could plan using the decoded grid, so the claim that the model itself pre-encodes multi-step plans needs an extra control or a more cautious interpretation. These are addressable weaknesses rather than fatal ones—adding controls and re-running the probes would either settle or substantially soften the concern.

Who this is for: interpretability and AI-safety researchers who want a reusable template for connecting behavior to internal state. The behavioral findings are a useful reference on their own. The paper does deserve a serious referee—the framework is worth engaging with and the central question is important. But my recommendation would be conditional acceptance at best: the authors should add the missing baselines and controls, release the code and data, and either provide evidence against the input-copy shortcut or pull back the causal reading of the decoded-map evaluation.

Referee Report

4 major / 4 minor

Summary. The paper proposes a framework that combines behavioural evaluation with representation probing to assess goal-directedness of an LLM agent in a grid-world navigation task. Behaviourally, it compares the agent's actions to A*-optimal policies across grid sizes, obstacle densities, iso-difficulty transformations, and instrumental/implicit-goal variants. Representationally, it trains linear and MLP probes on last-token activations, augmented with queried-cell (x, y) coordinates, to decode cell types ('cognitive maps'); it then checks whether the agent's actions are optimal with respect to its decoded map, and trains a one-shot transformer decoder to read multi-step action plans from activations. The paper concludes that the agent non-linearly encodes a coarse spatial map preserving task-relevant cues, that its actions are broadly consistent with this map, and that reasoning reorganises representations from broader spatial/long-horizon information toward immediate action selection. The authors argue that these findings support the need for introspective, representation-based evaluation beyond behaviour.

Significance. If the representational results were secured, the paper would make a useful contribution to goal-directedness evaluation: the behavioural protocols are carefully designed, the iso-difficulty and instrumental-goal experiments are informative, the Wilcoxon signed-rank tests are appropriate, and the one-shot plan decoder is a thoughtful attempt to avoid probe-created autoregressive structure. The behavioural findings—especially the attraction to a reward-irrelevant key and robustness to iso-difficulty transformations—are credible and interesting. However, the paper's central novelty, the claim that internal representations add information beyond behaviour, rests on probe results that currently lack essential controls. The cognitive-map probes are trained on activations from a model that literally sees the full grid text, yet no baselines are given for coordinate-prior, input-copy, or majority-class shortcuts. The plan-decoder comparison is only against random guessing. Consequently, the significance is conditional on the representational claims surviving appropriate controls.

major comments (4)

[§5.1] The cognitive-map probe is trained by concatenating activations with (x, y) coordinate features, and the model's input context contains the full grid text. No control probes are reported. For small grids padded to 15×15, the padding class covers 176/225 ≈ 78% of cells for a 7×7 grid, so a trivial all-padding predictor already achieves ~78% accuracy; the reported ~70% overall accuracy is therefore uninterpretable without a per-size majority baseline. More broadly, high accuracy could reflect coordinate-prior or input-copy shortcuts rather than an internal 'belief' or 'cognitive map.' The authors should add: majority-class baselines per grid size, coordinate-only probes, probes on random/shuffled activations, and an ablation that hides or scrambles the grid input. This is required to support the abstract's claim of a non-linear coarse spatial map.
[§5.1–§5.3] It is never stated whether probe training and test examples are split by grid instance. If activations from the same grid configuration appear in both training and test sets, the probe can memorize grid-specific features, inflating all decoding accuracies and the §5.2 policy-consistency metrics. Please state the instance-level split; if none exists, rerun with disjoint grids.
[§5.2] The Recovery metric and the conclusion that 'a substantial fraction of failures can be attributed to inaccurate, or fuzzy world representations' assume that the decoded grid is a valid estimate of the agent's internal beliefs. Because the probe lacks the controls above and is trained on ground-truth labels from the same observably available grid, the decoded map may simply mirror the input text. The Rec. metric then measures consistency with the input, not with internal beliefs. The authors need an independent validation of the probe (e.g., intervention on the decoded map, or consistency under input perturbations that change the ground truth) before this conclusion can be drawn.
[§5.3] Plan decoding is only compared against random chance (0.25^N). Since the target action sequence is the executed trajectory of the same model, and the full grid is in the prompt, a probe could achieve above-random prefix accuracy by learning a direct input-to-action policy rather than decoding a multi-step plan. The one-shot decoder rules out autoregressive shortcuts, but not this input-copy shortcut. Please add a control using activations from a model that never received the grid, or from representations where the grid text is permuted, and report whether prefix accuracy above baseline persists.

minor comments (4)

[§4.2] Table 3 reports eight Wilcoxon signed-rank tests without correction for multiple comparisons. Adjusting p-values or reporting them as exploratory would help; the reported effect sizes are appreciated.
[Fig. 6] No confidence intervals or significance tests are reported for probe accuracy differences (e.g., MLP vs. linear, 7×7 vs. 11×11). Given the small number of grids, bootstrap confidence intervals would be appropriate.
[App. A] Appendix A describes planned partially-observable experiments in the present/future tense ('we evaluate...', 'we ideally are able...') but these were not carried out. Please mark this clearly as a proposal rather than completed analysis.
[§5.2, footnote 6] The footnote acknowledges that decoded maps may reflect multiple nearby cells rather than a single location. This undercuts the single-cell decoding used for Acc-Dec and Rec. Consider deriving a distribution over policies to account for the probe's uncertainty.

Circularity Check

0 steps flagged

No significant circularity: the paper's claims are empirical measurements with separate train/test splits and independently recomputed optimal policies.

full rationale

The paper is an empirical evaluation rather than a derivation chain, and I found no step where a predicted quantity reduces by construction to a fitted input or to a self-citation. Behavioural metrics (§4) are computed by comparing GPT-OSS-20B's actions to A*-derived optimal policies on the same grids; these are measurements, not predictions fitted to the conclusion. The cognitive-map probes (§5.1) are trained on ground-truth cell labels from a train/test split, and the 'decoded cognitive maps' are then used in §5.2 to recompute optimal policies with A* independently of the agent's actions, so the 'actions are broadly consistent with internal representations' claim is a measured correlation rather than an identity. The plan decoder (§5.3) is trained on held-out executed trajectories, uses one-shot decoding to prevent the probe itself from constructing plan structure autoregressively, and is compared against a 0.25^N random baseline; no equation in the paper equates the decoded plan to the training labels by construction. The references to prior probing work (Li et al. 2023; Nanda et al. 2023; Gurnee & Tegmark 2024) are external evidence and are not load-bearing self-citations; I found no uniqueness theorem or ansatz smuggled in via the authors' own prior work. The coordinate-augmented probing procedure could in principle allow the probe to exploit coordinate priors or a near-literal copy of the grid text in the input context, and the absence of control probes is a legitimate validity and robustness concern; however, that is a correctness/experimental-control issue, not circularity under the criteria here. The paper itself acknowledges that establishing causal links is future work, which further supports treating the representational findings as empirical rather than definitional.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 0 invented entities

The paper rests mostly on standard ML/interpretability assumptions and experimental settings rather than new physical or mathematical axioms. The risky assumptions are about the faithfulness of supervised probes as readouts of model internals, and about whether chosen hyperparameters (low reasoning, horizon cap, probe layers) distort the conclusions.

free parameters (5)

Horizon cap factor c=1.5 = 1.5 × optimal path length L
Trajectories are truncated at 1.5×L to filter backtracking episodes; this hand-chosen threshold shapes all behavioural metrics.
Reasoning effort = low
Set to low because higher reasoning effort times out on large/dense grids; this choice affects every behavioural and representational result.
Probe layers = {7, 15, 23}
Activations are extracted from layers 7, 15, and 23; layer 15 is used for the main cognitive-map results. The selection is manual.
Plan decoding horizon T = 10
The plan decoder predicts 10 action steps; the reported prefix-accuracy curves depend on this fixed horizon.
Probe hidden dimension = 1024
MLP hidden dimension and decoder bottleneck size; standard capacity choice, not fitted to data but hand-chosen.

axioms (5)

standard math A* with Manhattan distance yields an optimal policy in a 4-directional grid world with obstacles.
Used in §4 to define the optimal reference policy; the heuristic is admissible and consistent for this transition model.
domain assumption The text-grid representation gives the LLM full observability of the true state, so a ground-truth optimal policy is computable and belief updating is unnecessary.
§3; this is the control that lets the authors attribute behavioral deviations to goal-directedness rather than perception.
domain assumption Residual-stream activations at a few chat-template tokens are a sufficient substrate for probing the model's environment representations and plans.
§5.1 and §5.3; a standard interpretability assumption, but not independently validated for this model and task.
domain assumption Iso-difficulty transformations preserve task difficulty (optimal path length), so any behavioral difference is attributable to task-irrelevant grid configuration.
§4.2; relies on geometric symmetries of reflected, rotated, transposed, and start-goal-swapped grids.
domain assumption Supervised probing with coordinate-conditioned inputs recovers the model's internal 'beliefs' rather than probe artifacts or input-text leakage.
§5.1; underpins the cognitive-map interpretation and is not validated with control probes or ablations.

pith-pipeline@v1.3.0-alltime-deepseek · 26057 in / 14632 out tokens · 133696 ms · 2026-08-03T03:04:46.025473+00:00 · methodology

0 comments

read the original abstract

Understanding an agent's goals helps explain and predict its behaviour, yet there is no established methodology for reliably attributing goals to agentic systems. We propose a framework for evaluating goal-directedness that integrates behavioural evaluation with interpretability-based analyses of models' internal representations. As a case study, we examine an LLM agent navigating a 2D grid world towards a goal state. Behaviourally, we evaluate the agent against optimal policies across varying grid sizes, obstacle densities, and goal structures, finding that performance scales with task difficulty while remaining robust to difficulty-preserving transformations and multi-goal structures. We then use probing methods to decode internal representations of the environment and multi-step action plans. We find that the LLM agent non-linearly encodes a coarse spatial map, preserving approximate task-relevant cues about its position and the goal location; that its actions are broadly consistent with these internal representations; and that reasoning reorganises them, shifting from spatial cues towards immediate action selection. Our findings support the view that introspective examination is required beyond behavioural evaluations to characterise how agents represent and pursue their objectives.

Figures

Figures reproduced from arXiv: 2602.08964 by Angelos Nalmpantis, Calum McNamara, Evgenii Kortukov, Fade Chen, Gabriele Sarti, Mario Giulianelli, Moksh Nirvaan, Niall Dalton, Raghu Arghal.

**Figure 1.** Figure 1: Overview of our goal-directedness analysis. A: We evaluate how iso-difficulty transforms affect agent trajectories that agree or disagree with the optimal policy. B: We prompt an LLMbased agent to reason and act over the fully-observable grid setup, extracting its pre-and post-reasoning activations at intermediate layers. C: We probe the agent’s beliefs over goal distance, planned actions and reconstruct… view at source ↗

**Figure 2.** Figure 2: Grid worlds with increasing wall density d, from fully open grids (d = 0) to maze-like grids with no circular paths (d = 1). natural text (Gurnee & Tegmark, 2024). Recent work has also probed LLMs for goal-oriented abstractions (Li et al., 2024) and shown that models engage in forward planning, pre-selecting future outputs before generating intermediate tokens (Pal et al., 2023; Men et al., 2024; Lindsey e… view at source ↗

**Figure 3.** Figure 3: An example grid (left) and its corresponding text based representation (right) used for LLM prompting. directly observes the true world state. This eliminates the need to maintain beliefs over hidden world states and allows optimal policies to be derived using standard algorithms. (2) Full observability also removes several factors that might otherwise confound the analysis, including memory, belief updat… view at source ↗

**Figure 4.** Figure 4: Top: Action accuracy (left) and mean JSD (right) in relation to the agent’s distance from the goal. Bottom: Action accuracy by size, complexity, and goal distance. using A* with Manhattan distance to the goal. This analysis assesses how closely the agent’s action choices and action distributions align with optimal behaviour in the grid world, without relying on or inspecting the agent’s internal represen… view at source ↗

**Figure 5.** Figure 5: Grid world variants with instrumental and implicit goals. In the text representation, the key and the door are encoded with K and D, and their meaning is explained in the system prompt. 4.3. Instrumental and Implicit Goals We move to examine whether the observed robustness extends to more complex goal structures using three variants of the grid world environment that include instrumental and implicit goal… view at source ↗

**Figure 6.** Figure 6: Extracting a cognitive map from GPT-OSS-20B representations. Left: Overall accuracy of an MLP and a linear probe. Center: Per-class recall (=accuracy) and precision for varying grid sizes. Right: A cognitive map decoded from pre-reasoning activations. of the agent’s actions with respect to its decoded, subjective cognitive map. Finally, in §5.3, we examine whether goal-directed action plans can be extract… view at source ↗

**Figure 7.** Figure 7: Probes performance for locating agent and goal positions. Binary localisation accuracy drops as the grid size increases, but avg. Manhattan distance to true locations remains bounded. structed by linear and MLP probes across various grid sizes. The MLP probe decodes cell identities with around 70% accuracy, reaching a maximum of 75.7% for 11 × 11 grids. Linear probes underperform at 39.1% average accuracy … view at source ↗

**Figure 8.** Figure 8: Performance of cognitive map before and after reasoning. Cognitive map accuracy drops significantly after reasoning. lower than 2 even for large 15 × 15 grids (see [PITH_FULL_IMAGE:figures/full_fig_p007_8.png] view at source ↗

**Figure 9.** Figure 9: Prefix accuracy of one-shot plan decoding before vs. after reasoning. We report the fraction of episodes for which the first N predicted actions match the target action sequence prefix. The dashed curve shows the random baseline 0.25N . For a predicted plan aˆ1:T and target action sequence a1:T (with T = 10), prefix accuracy at N is Pr[aˆ1:N = a1:N ]. We report prefix accuracy for N ∈ {1, . . . , 10} for b… view at source ↗

**Figure 10.** Figure 10: Partially Observable representation of a grid. The agent is at "A" and the goal (unseen) will be represented by "G". Hidden spaces are represented by "*", while revealed spaces are represented by "_". This setting is formalised by the Partially Observable Markov Decision Process (POMDP). A POMDP is a 7- tuple (S, A, T , R, O, Z, γ), which includes all elements of an MDP plus: • O: A finite set of observa… view at source ↗

**Figure 13.** Figure 13: Robustness to iso-difficulty environment transformations [PITH_FULL_IMAGE:figures/full_fig_p015_13.png] view at source ↗

**Figure 11.** Figure 11: Performance metrics by size complexity. (a) Goal success rate, (b) Policy entropy, (c) JSD from optimal policy, and (d) Expected Calibration Error. D. Iso-difficulty Transform Quantitative Results Original RotateEnv ReflectEnv StartGoalSwap TransposeEnv [PITH_FULL_IMAGE:figures/full_fig_p015_11.png] view at source ↗

**Figure 12.** Figure 12: Examples of iso-difficulty transformations. ReflectEnv RotateEnv StartGoalSwap TransposeEnv 0.5 0.0 0.5 Goal Success Goal Success (vs Baseline) ReflectEnv RotateEnv StartGoalSwap TransposeEnv 0.5 0.0 0.5 Accuracy Accuracy (vs Baseline) [PITH_FULL_IMAGE:figures/full_fig_p015_12.png] view at source ↗

**Figure 15.** Figure 15: Performance of size-specific MLP probes. Size-specific probes provide comparable performance, but less flexibility than the size-agnostic approach [PITH_FULL_IMAGE:figures/full_fig_p017_15.png] view at source ↗

discussion (0)

Reference graph

Works this paper leans on

8 extracted references · 1 canonical work pages

[3]

Edman, L., Schmid, H., and Fraser, A

URL https://openreview.net/forum? id=Ce79P8ULPY. Edman, L., Schmid, H., and Fraser, A. CUTE: Mea- suring LLMs’ understanding of their tokens. In Al- Onaizan, Y ., Bansal, M., and Chen, Y .-N. (eds.),Pro- ceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pp. 3017–3026, Miami, Florida, USA, November 2024. Association for C...

doi:10.18653/v1/2024.emnlp-main 2024
[37]

conll-1.37/

URL https://aclanthology.org/2023. conll-1.37/. Rajcic, N. and Søgaard, A. Goal-Directedness is in the Eye of the Beholder, August 2025. Savage, L. J. Samuelson’s Foundations: Its Mathemat- ics.Journal of Political Economy, 56(3):200–202, June 1948. ISSN 0022-3808. doi: 10.1086/256672. URLhttps://www.journals.uchicago.edu/ doi/abs/10.1086/256672. Publishe...

work page doi:10.1086/256672 2023
[177]

emnlp-main.177/

URL https://aclanthology.org/2024. emnlp-main.177/. Everitt, T., Garbacea, C., Bellot, A., Richens, J., Papadatos, H., Campos, S., and Shah, R. Evaluating the goal- directedness of large language models, 2025. URL https://arxiv.org/abs/2504.11844. Goldstein, S. and Lederman, H. What Does ChatGPT Want? An Interpretationist Guide, September 2025. URL https:...

Pith/arXiv arXiv 2024
[440]

emnlp-main.440/

URL https://aclanthology.org/2024. emnlp-main.440/. Naik, A., Quinn, P., Bosch, G., Gouné, E., Zabala, F. J. C., Brown, J. R., and Young, E. J. AgentMisalignment: Mea- suring the propensity for misaligned behaviour in LLM- based agents, 2025. URL https://arxiv.org/ abs/2506.04018. Nanda, N., Lee, A., and Wattenberg, M. Emergent lin- ear representations in...

Pith/arXiv arXiv 2024
[1434]

emnlp-main.1434/

URL https://aclanthology.org/2025. emnlp-main.1434/. Davidson, D. Radical interpretation.Dialectica, pp. 313– 328, 1973. doi: 10.1111/j.1746-8361.1973.tb00623. x. URL https://www.jstor.org/stable/ 42968535. Publisher: JSTOR. Dennett, D. C. The Interpretation of Texts, People and Other Artifacts.Philosophy and Phenomenological Re- search, 50:177–194, 1990....

arXiv 2025
[1593]

scheming

URL https://www.science.org/doi/ abs/10.1126/science.275.5306.1593. Skalse, J. and Abate, A. Misspecification in Inverse Re- inforcement Learning, March 2023. URL http:// arxiv.org/abs/2212.03201. arXiv:2212.03201 [cs]. Summerfield, C., Luettgau, L., Dubois, M., Kirk, H. R., Hackenburg, K., Fist, C., Slama, K., Ding, N., Ansel- metti, R., Strait, A., et a...

arXiv 2023
[2024]

URL https://openreview.net/forum? id=fh8EYKFKns. OpenAI. gpt-oss-120b & gpt-oss-20b model card, 2025. URLhttps://arxiv.org/abs/2508.10925. Pal, K., Sun, J., Yuan, A., Wallace, B., and Bau, D. Fu- ture lens: Anticipating subsequent tokens from a sin- gle hidden state. In Jiang, J., Reitter, D., and Deng, S. (eds.),Proceedings of the 27th Conference on Comp...

Pith/arXiv arXiv 2025
[2025]

Chalmers, D

URL https://openreview.net/forum? id=DzGe40glxs. Chalmers, D. J. Propositional interpretability in artificial in- telligence, 2025. URL https://arxiv.org/abs/ 2501.15740. Chevalier-Boisvert, M., Dai, B., Towers, M., Perez-Vicente, R., Willems, L., Lahlou, S., Pal, S., Castro, P. S., and Terry, J. Minigrid & miniworld: Modular & customizable reinforcement ...

Pith/arXiv arXiv 2025

[1] [3]

Edman, L., Schmid, H., and Fraser, A

URL https://openreview.net/forum? id=Ce79P8ULPY. Edman, L., Schmid, H., and Fraser, A. CUTE: Mea- suring LLMs’ understanding of their tokens. In Al- Onaizan, Y ., Bansal, M., and Chen, Y .-N. (eds.),Pro- ceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pp. 3017–3026, Miami, Florida, USA, November 2024. Association for C...

doi:10.18653/v1/2024.emnlp-main 2024

[2] [37]

conll-1.37/

URL https://aclanthology.org/2023. conll-1.37/. Rajcic, N. and Søgaard, A. Goal-Directedness is in the Eye of the Beholder, August 2025. Savage, L. J. Samuelson’s Foundations: Its Mathemat- ics.Journal of Political Economy, 56(3):200–202, June 1948. ISSN 0022-3808. doi: 10.1086/256672. URLhttps://www.journals.uchicago.edu/ doi/abs/10.1086/256672. Publishe...

work page doi:10.1086/256672 2023

[3] [177]

emnlp-main.177/

URL https://aclanthology.org/2024. emnlp-main.177/. Everitt, T., Garbacea, C., Bellot, A., Richens, J., Papadatos, H., Campos, S., and Shah, R. Evaluating the goal- directedness of large language models, 2025. URL https://arxiv.org/abs/2504.11844. Goldstein, S. and Lederman, H. What Does ChatGPT Want? An Interpretationist Guide, September 2025. URL https:...

Pith/arXiv arXiv 2024

[4] [440]

emnlp-main.440/

URL https://aclanthology.org/2024. emnlp-main.440/. Naik, A., Quinn, P., Bosch, G., Gouné, E., Zabala, F. J. C., Brown, J. R., and Young, E. J. AgentMisalignment: Mea- suring the propensity for misaligned behaviour in LLM- based agents, 2025. URL https://arxiv.org/ abs/2506.04018. Nanda, N., Lee, A., and Wattenberg, M. Emergent lin- ear representations in...

Pith/arXiv arXiv 2024

[5] [1434]

emnlp-main.1434/

URL https://aclanthology.org/2025. emnlp-main.1434/. Davidson, D. Radical interpretation.Dialectica, pp. 313– 328, 1973. doi: 10.1111/j.1746-8361.1973.tb00623. x. URL https://www.jstor.org/stable/ 42968535. Publisher: JSTOR. Dennett, D. C. The Interpretation of Texts, People and Other Artifacts.Philosophy and Phenomenological Re- search, 50:177–194, 1990....

arXiv 2025

[6] [1593]

scheming

URL https://www.science.org/doi/ abs/10.1126/science.275.5306.1593. Skalse, J. and Abate, A. Misspecification in Inverse Re- inforcement Learning, March 2023. URL http:// arxiv.org/abs/2212.03201. arXiv:2212.03201 [cs]. Summerfield, C., Luettgau, L., Dubois, M., Kirk, H. R., Hackenburg, K., Fist, C., Slama, K., Ding, N., Ansel- metti, R., Strait, A., et a...

arXiv 2023

[7] [2024]

URL https://openreview.net/forum? id=fh8EYKFKns. OpenAI. gpt-oss-120b & gpt-oss-20b model card, 2025. URLhttps://arxiv.org/abs/2508.10925. Pal, K., Sun, J., Yuan, A., Wallace, B., and Bau, D. Fu- ture lens: Anticipating subsequent tokens from a sin- gle hidden state. In Jiang, J., Reitter, D., and Deng, S. (eds.),Proceedings of the 27th Conference on Comp...

Pith/arXiv arXiv 2025

[8] [2025]

Chalmers, D

URL https://openreview.net/forum? id=DzGe40glxs. Chalmers, D. J. Propositional interpretability in artificial in- telligence, 2025. URL https://arxiv.org/abs/ 2501.15740. Chevalier-Boisvert, M., Dai, B., Towers, M., Perez-Vicente, R., Willems, L., Lahlou, S., Pal, S., Castro, P. S., and Terry, J. Minigrid & miniworld: Modular & customizable reinforcement ...

Pith/arXiv arXiv 2025