The Misclassification of Autistic Writing as AI-Generated
Research finds AI detection models frequently misclassify autistic writing as AI-generated, exhibiting bias against minority groups.
arXiv cs.CL — Computation and Language·17 Jul
Search signals, briefings, company results, benchmarks and glossary terms.
Search signals, briefings, company results, benchmarks and glossary terms.
Use this view to inspect the underlying evidence corpus. For ranked developments, decision posture and interpretation, use Signals.
Raw feed or Signals?
Raw feed is chronological evidence. Signals ranks and interprets material change.
Research finds AI detection models frequently misclassify autistic writing as AI-generated, exhibiting bias against minority groups.
A research paper studies memory-augmented conversational agents across 24 participants and 10 sessions, assessing relational constructs like self-disclosure and perceived memory.
Research explores using mask-aware policy gradients to apply reinforcement learning more effectively to Masked Diffusion Language Models (MDLMs).
Research introduces Polestar, a method for improving diffusion LLM inference efficiency by addressing bidirectional attention and KV-cache reuse issues.
A research paper analyzes 109 Qubes Security Bulletins (2011–2025) and Xen Security Advisories to measure public security records.
Research introduces PRISM, a complex-valued neural network architecture exploring semantic phase locking and interference to disentangle semantic importance from activation magnitude.
Researchers propose Decoupled Alignment, a training-free method using knowledge distillation to enhance LLM safety and prevent 'shadow alignment' during adaptation.
Research explores using language models (LMs) as evaluators for other LMs, enhancing evaluation quality by increasing compute during test-time reasoning.
Research explores evolving LLM evaluation rubrics from a single query using synthetic pairwise evidence, enhancing fine-grained model assessment.
SAMark proposes a self-anchored text watermarking method robust against paragraph-level paraphrasing by removing sentence order dependency.
Research explores automatically evolving task-specific prompt guidelines to address underspecified user queries and improve LLM reliability.
LLM evaluators (reward models and LLM-as-a-Judge) exhibit significant language bias, failing to assign consistent scores across 23 languages for semantically identical instruction-response pairs.
Research investigates LLM capability to translate conceptual research ideas into structured plans, demonstrating potential for scientific discovery acceleration.
Research explores augmenting natural language interfaces for data exploration with semantic and context-aware query recommendations over multi-table relational databases.
A study found single-pass frontier models outperform multi-agent debate tools for providing useful feedback on economics meta-analyses.
Research analyzed 4,500 LLM-chatbot interactions to model user engagement dynamics in English as a Foreign Language (EFL) writing with "Penny."
Research introduces PUPPET, a taxonomy and resource to predict human belief change in dialogues with manipulative LLMs.
Research claims standard hate speech detectors over-attribute 'hate' by conflating it with partisan or geopolitical hostility in social media content.
Researchers propose "Verbalized Sampling" to mitigate mode collapse in LLMs, attributing it to annotator bias favoring familiar text.
Step-Tagging is a lightweight sentence-classifier framework designed to control and optimize Language Reasoning Model generation by reducing inefficient verification steps.
Research demonstrates that pretraining data can be poisoned through computational propaganda to induce harmful behaviors in large language models, even with data curation.
Research finds static retrieval evaluation methods fail to predict utility for multi-step LLM agents, which use documents to inform subsequent actions.
RetroAgent uses LLMs with structured memory for multi-step retrosynthesis planning, improving upon traditional tree search methods.
PERL is a new research model for Chinese Automatic Speech Recognition (ASR) error correction, addressing phonetic errors and length constraints.
Research explores integrating machine unlearning with RAG to manage sensitive information and harmful content retention in LLMs.
Research questions the consistent gains of tool use in web agents, citing limited experimental scales and non-comparable settings in prior studies.
Research proposes LLM agents can self-manage context via 'state proprioception' to overcome long-horizon context window limitations.
Research from arXiv explores how large language models, as a new communication medium, may transmit cultural patterns into human language.
Researchers propose new algorithms to segment human-LLM co-authored text, moving beyond binary classification to localize generated segments.
MonteRET is a new AI agent framework enhancing multimodal LLMs with multi-granularity knowledge retrieval for chest CT report generation.
Research describes a system combining speaker diarization with a Qwen3-ASR-1.7B model for multilingual two-speaker conversational speech.
Research reformulates Transformer/Attention mechanisms via measure theory and frequency analysis, claiming hallucination is an inevitable structural LLM limitation.
Research introduces "tool efficiency" and "marginal tool utility" as new quantitative metrics to evaluate the rate and usefulness of LLM agent tool calls.
A research paper introduces ArogyaSutra, a multi-agent framework for multimodal medical reasoning in Indic languages, targeting low-resource healthcare.
Research identifies 'semantic register compression' as a failure mode in multi-agent LLM systems, where intermediate agents lose critical semantic distinctions.
Research investigates if neural language models distinguish grammatical from ungrammatical strings, moving beyond probability-based metrics.
Research developed and validated the Generative AI Reliance Types Scale (GenAI-RTS) to measure how students rely on generative AI in academic writing.
New research introduces LBA, a method for generating high-quality adversarial texts with significantly lower query budgets in hard-label scenarios.
Researchers propose MemTrace, a novel framework for tracing and attributing errors in large language model memory systems to improve reliability.
FlowBot proposes using bilevel optimization and textual gradients to automatically induce LLM workflows, moving beyond human-crafted pipelines.
Research introduces persona vectors to systematically audit open-weight LLMs across 53 behavioral traits, revealing suppressed or resistant behaviors.
Research on Gemma-3-27B-it reveals first-language (L1) scoring bias in LLM-based automated essay scoring, impacting cross-prompt generalization.
WrAFT, an automated writing evaluation system, uses LLMs like LLaMA-3.3-70B-Instruct and GPT-4o for modular scoring and feedback.
Research introduces ShopX, a foundation model designed for agentic shopping to directly fulfill complex intent-to-item needs, bypassing traditional search.
Federated learning on clinical text can leak sensitive data via gradient inversion; tokenizer choice impacts privacy risk.
Research paper proposes an auditable single-system method to evaluate LLM honesty by using a game engine as ground truth, not model self-assessments.
Research benchmarks six MLLMs on a scientific visualization literacy test, finding current evaluations are chart-centric and lack SciVis understanding evidence.
Research explores instruction tuning and model merging to adapt reasoning language models for domains lacking reliable output verification.
Researchers introduced Just Keep Prompting (JKP), a new multi-turn evaluation framework for Vision-Language Models (VLMs) to test stability under sustained questioning.
Research introduces Nous, a belief-based memory architecture for LLM agents using Bayesian inference and information theory for belief revision and forgetting.
© 2026 OneBench: AI Insights. All rights reserved.
Evidence before opinion