The Archive

Multi-directional Perturbation-Rewrite Search framework uses LLMs to automate heuristic design for NP-hard combinatorial optimization with factorized transformations.

LLMs enable formal software verification at scale, reducing manual proof engineering burden for systems assurance.

LongZero uses RL from self-generated tasks to improve LLM long-context reasoning without external training data or teacher models.

Environment Duels benchmark evaluates frontier LLMs on creating and solving interactive environments; reveals distinct authoring vs. solving abilities.

P2NPCBench pairs NP-complete and polynomial-time problems via AgentGAN adversarial rewriting to test LLM reasoning beyond recall on complexity boundaries.

Code Comprehensor uses dependency callgraphs and hierarchical documentation for large repository code understanding by LLMs.

Agent Behavior Maps profile LLM-harness combinations via action log-likelihoods on reference trajectories, enabling performance prediction across 100 LLMs.

System One decision models for legal authority verification with abstention, tested on CaseHOLD dataset with missing holdings.

As large language models (LLMs) continue to grow in scope and capability, privacy concerns are on the rise. Since users increasingly interact with a single general-purpose system across many distinct contexts, there is a natural tendency to pursue a one-size-fits-all solution to privacy. Unfortunately, this overlooks a crucial insight from decades of privacy research: privacy is inherently contextual. In this paper, we argue that existing work in prompt privacy is drifting towards attempts to design a unified framework, which is highly misleading and can foster a false sense of security. Furt...

As LLMs are increasingly deployed in value-sensitive contexts, aligning them with human values has emerged as a post-training objective in NLP, aimed at instilling additional values and guiding model behaviour. However, current evaluations remain narrow. Because values are interconnected, it is unclear how fine-tuning for specific values alters unrelated capabilities and behaviours. In this work, we adapt four instruction-tuned LLMs across individual Hofstede's value dimensions and evaluate them on both cross-value transfer and broader behaviours. We find that value-specific training produces...

Empirical study comparing LLM-based and classical feature selection on tabular data with GRPO-adapted Qwen and Llama models.

OptionWorld benchmark reveals LLMs fail at multi-day task completion, with 35.5% of failures due to lost context before deadline.

Analysis of LLM generic acceptability judgments aligned with psycholinguistic factors across twelve models.

Polarity blindness vulnerability: LLMs fail safety checks after negation and antonym substitution despite preserved semantic meaning.

CollabSciReM aligns LLMs for scientific collaboration via science-aware reward decomposition, balancing accuracy, rigor, and token efficiency across multi-turn interactions.

Benchmark study of associative creativity in MLLMs vs. humans on abstract visual stimuli with controlled naming difficulty; evaluates originality and effectiveness.

LLMs hallucinate in cross-language code translation via semantic conflicts, target-language violations, and inconsistent dependencies; existing mitigations insufficient.

COMPASS agentic router selects optimal LLMs by profiling capabilities via tools and pairwise selection, achieving 69.85% accuracy at lower cost than trained baselines.

Mechanistic interpretability study investigates circuit overlap between linguistic and arithmetic capabilities in LLMs using causal analysis.

OpenVLHarness framework enables multimodal LLMs to use external tools and retain evidence across reasoning steps for visual agentic tasks.

LatComm introduces reusable latent context compression between LLMs, enabling compact encoding that is readable across multiple models and questions.

Game-theoretic framework enables LLMs to learn scam-baiting strategies through in-context learning without fine-tuning.

TokenGate enables learnable token pruning in LLMs for faster, memory-efficient long-context prefill using a differentiable top-K approximation.

The vibecoding tag has been getting more and more useless for a while now. It lost the original meaning as the culture of software engineering with LLMs matured a bit. It got silly when posts talking about LLMs in general got tagged. It got useless when posts even mentioning some project created with LLMs in passing got tagged. Now it actually got harmful to the community when tagging posts turns away very valuable members https://lobste.rs/c/vecuxj I sighed and moved on for a while now as almost everything gets tagged - you can skip a word. But now it not only carries no signal whatsoever, i...

Retrieval Program Learning enables LLMs to compose search primitives via RL into adaptive multi-step retrieval programs with state and control flow.

MisleadQA diagnostic dataset reveals LLMs fail on out-of-scope but factually true context, not just false or irrelevant information.

I'm curious what stance others have taken? Or what they've experienced with this? I'm toying with adding a rule to our values.md that the LLM can only significantly generate one or the other. I'm reviewing a merge request at the moment where it's clear the LLM mostly did both, and they're highly tautological. And I can see how it over engineered the code, just to be able to write more tests almost.

Paired-residual subspace removal improves multilingual text embeddings from frozen LLMs by disentangling language variation.

Mental World Model framework for simulating subjective cognition and Theory of Mind extends LLMs beyond linguistic inference with dynamic evolution.

VISTA infers latent user values from survey responses and uses hypernetworks to steer frozen LLMs toward personalized outputs without prompt engineering.