Frontier Coding Agents Use Metaprogramming to Adapt to Unfamiliar Programming Languages

LLM-based coding agents are usually evaluated in familiar software settings: mainstream languages, common libraries, and public repositories. These benchmarks remain important, but they can hide how agents behave when the language itself is unfamiliar. We evaluate six contemporary coding agents on four esoteric programming languages using a sequential setup with file editing, local execution, and hidden-test grading. Our protocol exposes capability differences between these agents that mainstream coding and agentic benchmarks such as SWE-Bench Verified and Terminal-Bench 2.0 compress into much narrower bands. We observe that the strongest agents, Claude Opus 4.6 and GPT-5.4 xhigh, often avoid writing the target language directly. On Brainfuck and Befunge-98, they write Python programs that generate target-language code and debug those generators locally. Forbidding this metaprogramming strategy causes large performance drops. Text guidance distilled from this strategy does not materially improve weaker agents. In contrast, Opus-derived Python helper code for building generators, with no solved benchmark programs or hidden-test answers, sharply improves Sonnet 4.6 and GPT-5.4 mini on the same problems, while Haiku 4.5 remains low. More interpreter calls and output tokens improve stronger agents but leave weaker agents near their original performance, indicating that these resources amplify useful strategies rather than create them. Together, these results show that strong coding agents adapt to unfamiliar languages by using tools, feedback, and workspace state to build a working model of the target language. Metaprogramming is the clearest case, but the broader gap is constructing and debugging a strategy that works under the target language's rules.

Do Vision-Language Models See or Guess? Measuring and Reducing Textual-Prior Reliance with a Phrasing-Controlled Benchmark

Vision-language models (VLMs) are increasingly deployed where answers must follow from what is in the image, yet they often answer from textual priors, the question's phrasing together with memorized world knowledge, rather than from the image itself, which inflates benchmark scores and yields confident but ungrounded answers. Existing benchmarks rarely isolate this behavior, since each image is usually paired with a single fixed question. To measure the reliance, we build a 540-image benchmark across six reasoning categories and generate four question variants over the same images, so that phrasing rather than image content is the controlled variable. The hardest variant is written directly from the image to minimize text leakage. We benchmark eleven VLMs spanning small open-weight models to large closed-source systems: every model degrades on the hardest variant, and open models fall furthest. Our central diagnostic is a no-image ablation, which collapses the open-weight models to their text-only floor (1 to 9 percent). Three further analyses, LLM-rated difficulty, low base-to-final textual similarity, and human re-annotation, corroborate genuine image-dependence. In-context exemplars that match how a variant was built recover the most accuracy, and GRPO post-training of a small VLM yields consistent gains across all four variants that transfer to a held-out out-of-distribution set. Textual-prior reliance is measurable and partly trainable away.

Geometry of Human Perceptual Domains Emerges Transiently in LLM Representations

While large language models (LLMs) are trained purely on textual data, prior work has shown that their internal representations can exhibit rich geometric structure in embedding space. Building on this line of work, we investigate whether such structure is similar to human perceptual organisation across different domains (e.g., color, pitch, emotion, and taste). Specifically, we study the layer-wise emergence of intrinsic geometrical structure corresponding to perceptual modalities within the residual streams of multiple open-weight transformer architectures. Our results reveal three key findings. First, we observe the emergence of layer-wise geometric structure across multiple perceptual domains, despite the absence of any direct perceptual supervision during training. Second, these perceptual domains exhibit distinct emergence profiles, with both geometric structure and its alignment with human baselines following domain- and model-specific trajectories across depth. Third, this emergence follows a consistent representational trajectory: geometry is weak or diffuse in early layers, becomes progressively organised in intermediate layers, and is attenuated in later layers, suggesting that perceptual geometry arises transiently as part of the model's internal transformation pipeline. This provides new insight into how and where human-like perceptual geometry arises in LLMs, offering a principled pathway for mechanistic analysis of internal representations.

Hybrid Neural World Models

Neural surrogates promise large speedups over classical solvers for physical dynamics but fail silently at sharp dynamical events such as shocks, fronts, and contact. We present hybrid neural world models for physical dynamics: a recipe for training and deploying multi-horizon surrogates in physical state space, where a single network with continuous horizon conditioning is trained with direct supervision against textbook reference solvers to predict any future state at horizon T in one forward pass. Although no part of the training data, loss function, or architecture supervises discontinuity location, the trained surrogate encodes it implicitly, recoverable from its forward passes alone as a per-trajectory error map that concentrates on shocks, fronts, and contacts, and stays small elsewhere. The map is competitive with or better than standard label-free baselines including deep ensembles, learned error heads, gradient-magnitude indicators, and locally-adaptive conformal prediction, while using only a single trained network and requiring no calibration set or governing-equation knowledge. The recipe supports two operating points. Mode 1 runs the surrogate alone for maximum throughput, with same-hardware CPU speedups of 26x to 72x against textbook solvers on the PDE environments. Mode 2 uses the error map to gate a reference-solver fallback, deferring uncertain trajectories and roughly halving the surrogate's residual error at the default operating point. The recipe applies without modification across reaction-diffusion, compressible Euler, and rigid-body collision dynamics.

Discovering Reinforcement Learning Interfaces with Large Language Models

Reinforcement learning systems rely on environment interfaces that specify observations and reward functions, yet constructing these interfaces for new tasks often requires substantial manual effort. While recent work has automated reward design using large language models (LLMs), these approaches assume fixed observations and do not address the broader challenge of synthesizing complete task interfaces. We study RL task interface discovery from raw simulator state, where both observation mappings and reward functions must be generated. We propose LIMEN (Code available at https://github.com/Lossfunk/LIMEN), a LLM guided evolutionary framework that produces candidate interfaces as executable programs and iteratively refines them using policy training feedback. Across novel discrete gridworld tasks and continuous control domains spanning locomotion and manipulation, joint evolution of observations and rewards discovers effective interfaces given only a trajectory-level success metric, while optimizing either component alone fails on at least one domain. These results demonstrate that automatic construction of RL interfaces from raw state can substantially reduce manual engineering and that observation and reward components often benefit from co-design, as single-component optimization fails catastrophically on at least one domain in our evaluation suite.

See, Symbolize, Act: Grounding VLMs with Spatial Representations for Better Gameplay

Vision-Language Models (VLMs) excel at describing visual scenes, yet struggle to translate perception into precise, grounded actions. We investigate whether providing VLMs with both the visual frame and the symbolic representation of the scene can improve their performance in interactive environments. We evaluate three state-of-the-art VLMs across Atari games, VizDoom, and AI2-THOR, comparing frame-only, frame with self-extracted symbols, frame with ground-truth symbols, and symbol-only pipelines. Our results indicate that all models benefit when the symbolic information is accurate. However, when VLMs extract symbols themselves, performance becomes dependent on model capability and scene complexity. We further investigate how accurately VLMs can extract symbolic information from visual inputs and how noise in these symbols affects decision-making and gameplay performance. Our findings reveal that symbolic grounding is beneficial in VLMs only when symbol extraction is reliable, and highlight perception quality as a central bottleneck for future VLM-based agents.

EsoLang-Bench: Evaluating Genuine Reasoning in Large Language Models via Esoteric Programming Languages

Large language models achieve near-ceiling performance on code generation benchmarks, yet these results increasingly reflect memorization rather than genuine reasoning. We introduce EsoLang-Bench, a benchmark using five esoteric programming languages (Brainfuck, Befunge-98, Whitespace, Unlambda, and Shakespeare) that lack benchmark gaming incentives due to their economic irrationality for pre-training. These languages require the same computational primitives as mainstream programming but have 1,000-100,000x fewer public repositories than Python (based on GitHub search counts). We evaluate five frontier models across five prompting strategies and find a dramatic capability gap: models achieving 85-95% on standard benchmarks score only 0-11% on equivalent esoteric tasks, with 0% accuracy beyond the Easy tier. Few-shot learning and self-reflection fail to improve performance, suggesting these techniques exploit training priors rather than enabling genuine learning. EsoLang-Bench provides the first benchmark designed to mimic human learning by acquiring new languages through documentation, interpreter feedback, and iterative experimentation, measuring transferable reasoning skills resistant to data contamination.

ISO-Bench: Can Coding Agents Optimize Real-World Inference Workloads?

We introduce ISO-Bench, a benchmark for coding agents to test their capabilities on real-world inference optimization tasks. These tasks were taken from vLLM and SGLang, two of the most popular LLM serving frameworks. Each task provides an agent with a codebase and bottleneck description, whereby the agent must produce an optimization patch evaluated against expert human solutions. We curated 54 tasks from merged pull requests with measurable performance improvements. While existing benchmarks heavily use runtime-based metrics, such approaches can be gamed to pass tests without capturing the actual intent of the code changes. Therefore, we combine both hard (execution-based) and soft (LLM-based) metrics to show that both are necessary for complete evaluation. While evaluating both closed and open-source coding agents, we find no single agent dominates across codebases. Surprisingly, agents often identify correct bottlenecks but fail to execute working solutions. We also show that agents with identical underlying models differ substantially, suggesting scaffolding is as important as the model.

Making Large Language Models Speak Tulu: Structured Prompting for an Extremely Low-Resource Language

Can large language models converse in languages virtually absent from their training data? We investigate this question through a case study on Tulu, a Dravidian language with over 2 million speakers but minimal digital presence. Rather than fine-tuning an LLM, we examine whether structured prompts alone can elicit basic conversational ability under controlled prompting. We systematically tackle various challenges posed by absence of training data for Tulu by combining explicit grammar documentation, negative constraints to suppress high-probability tokens from related languages, romanization standardization, and quality-controlled synthetic data generation via self-play. Evaluated on a manually curated held-out set across three LLMs (Gemini 2.0 Flash, GPT-4o, Llama 3.1 70B) and validated by native speakers, our approach reduces vocabulary contamination from 80% to 5% while achieving 85% grammatical accuracy. Cross-model analysis reveals that negative constraints provide consistent improvements (12--18 percentage points), while grammar documentation effects vary by model architecture (8--22 points).

Building Interpretable Models for Moral Decision-Making

We build a custom transformer model to study how neural networks make moral decisions on trolley-style dilemmas. The model processes structured scenarios using embeddings that encode who is affected, how many people, and which outcome they belong to. Our 2-layer architecture achieves 77% accuracy on Moral Machine data while remaining small enough for detailed analysis. We use different interpretability techniques to uncover how moral reasoning distributes across the network, demonstrating that biases localize to distinct computational stages among other findings.