A trace-based benchmark of 30 security tasks finds that less-restricted LLM derivatives outperform stock safety-aligned models on some agent tasks for Gemma but not Qwen or Llama, with similar patterns on non-security controls.
hub
Release Strategies and the Social Impacts of Language Models
29 Pith papers cite this work. Polarity classification is still indexing.
29
Pith papers citing it
abstract
Large language models have a range of beneficial uses: they can assist in prose, poetry, and programming; analyze dataset biases; and more. However, their flexibility and generative capabilities also raise misuse concerns. This report discusses OpenAI's work related to the release of its GPT-2 language model. It discusses staged release, which allows time between model releases to conduct risk and benefit analyses as model sizes increased. It also discusses ongoing partnership-based research and provides recommendations for better coordination and responsible publication in AI.
hub tools
citation-role summary
background 3
baseline 1
citation-polarity summary
representative citing papers
Adapts change point detection to segment human-LLM co-authored text using weighted and generalized algorithms with minimax optimality and strong empirical results against baselines.
Open source AI shows lower collaboration intensity, reduced direct contributions, and a shift toward adaptive use rather than joint improvement compared to traditional OSS.
ExaGPT uses span-level similarity retrieval from human and LLM datastores to detect machine-generated text while supplying the matching spans as human-interpretable evidence, achieving up to 37-point accuracy gains over prior interpretable detectors at 1% FPR.
LLMs trained on simple specification gaming generalize to zero-shot reward tampering including rewriting their own reward function.
Multitask fine-tuning of an encoder-decoder model on prompted datasets produces zero-shot generalization that often beats models up to 16 times larger on standard benchmarks.
RAG models set new state-of-the-art results on open-domain QA by retrieving Wikipedia passages and conditioning a generative model on them, while also producing more factual text than parametric baselines.
An image-semantic guided method enhances MLLMs for detecting AI-generated modern Chinese poetry by combining poem text with visual representations of content, achieving 85.65% Macro-F1 with Gemini and outperforming text baselines and RoBERTa.
MELD is a multi-task AI-text detector using auxiliary heads, uncertainty-weighted losses, EMA distillation, and pairwise ranking that reaches 99.9% TPR at 1% FPR on a new held-out benchmark while remaining competitive on the RAID leaderboard.
BREW achieves TPR of 0.965 and FPR of 0.02 under 10% synonym substitution by shifting from ECC decoding to designated verification with block voting and local validation.
DSIPA is a zero-shot black-box detector that uses sentiment distribution consistency and preservation metrics to identify LLM text, reporting up to 49.89% F1 gains over baselines across domains and models.
IRM derives implicit reward signals from off-the-shelf LLMs to detect generated text zero-shot and reports better results than prior zero-shot and supervised detectors on the DetectRL benchmark.
A proposed pipeline shows LLMs introduce detectable race and gender biases when summarizing life narratives, creating potential for representational harm in research.
A human-centered design workshop with journalism practitioners yields an evaluation cookbook and design requirements for contextualized, value-aligned generative AI benchmarks.
GigaCheck detects LLM-generated text at both document and span levels by combining fine-tuned language-model embeddings with a DETR-like architecture that treats generated intervals as detectable objects.
Recursive paraphrasing attacks substantially lower detection rates for multiple AI text detectors with only minor quality loss, while a theoretical analysis ties best-case AUROC to total variation distance between human and AI distributions.
The authors provide a detailed taxonomy of 21 risks associated with language models, covering discrimination, information leaks, misinformation, malicious applications, interaction harms, and societal impacts like job loss and environmental costs.
CodeXGLUE supplies a standardized collection of 10 code-related tasks, 14 datasets, an evaluation platform, and BERT-, GPT-, and encoder-decoder-style baselines.
Reveals hidden human-like spans in machine-generated texts that raise detection complexity and proposes a stacked enhancement framework with hard-EM optimization to improve detectors across LLMs.
A multi-level framework that models local and global relations among token detection scores to improve machine-generated text detection with low overhead.
DetectRL-X is a multilingual benchmark evaluating LLM text detectors on 8 languages, 6 domains, 4 commercial generators, and paraphrasing/perturbation attacks.
LAPD, derived from the provable preference discrepancy in aligned LLMs, improves zero-shot AI text detection by 45.82% over baselines with claimed statistical dominance over Fast-DetectGPT.
Genre and model exert stronger influence on writing style than human/LLM source or decoding strategy in a broad comparison of lexicogrammatical features.
A rate-distortion framework for lossy compression of transformer representations yields substantial bitrate savings on language tasks while preserving accuracy, with observed rates aligning to derived information-theoretic bounds.
citing papers explorer
-
Measuring Safety Alignment Effects in Autonomous Security Agents
A trace-based benchmark of 30 security tasks finds that less-restricted LLM derivatives outperform stock safety-aligned models on some agent tasks for Gemma but not Qwen or Llama, with similar patterns on non-security controls.
-
Segmenting Human-LLM Co-authored Text via Change Point Detection
Adapts change point detection to segment human-LLM co-authored text using weighted and generalized algorithms with minimax optimality and strong empirical results against baselines.
-
From OSS to Open Source AI: an Exploratory Study of Collaborative Development Paradigm Divergence
Open source AI shows lower collaboration intensity, reduced direct contributions, and a shift toward adaptive use rather than joint improvement compared to traditional OSS.
-
ExaGPT: Example-Based Machine-Generated Text Detection for Human Interpretability
ExaGPT uses span-level similarity retrieval from human and LLM datastores to detect machine-generated text while supplying the matching spans as human-interpretable evidence, achieving up to 37-point accuracy gains over prior interpretable detectors at 1% FPR.
-
Sycophancy to Subterfuge: Investigating Reward-Tampering in Large Language Models
LLMs trained on simple specification gaming generalize to zero-shot reward tampering including rewriting their own reward function.
-
Multitask Prompted Training Enables Zero-Shot Task Generalization
Multitask fine-tuning of an encoder-decoder model on prompted datasets produces zero-shot generalization that often beats models up to 16 times larger on standard benchmarks.
-
Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks
RAG models set new state-of-the-art results on open-domain QA by retrieving Wikipedia passages and conditioning a generative model on them, while also producing more factual text than parametric baselines.
-
Seeing the Poem: Image-Semantic Detection of AI-Generated Modern Chinese Poetry with MLLMs
An image-semantic guided method enhances MLLMs for detecting AI-generated modern Chinese poetry by combining poem text with visual representations of content, achieving 85.65% Macro-F1 with Gemini and outperforming text baselines and RoBERTa.
-
MELD: Multi-Task Equilibrated Learning Detector for AI-Generated Text
MELD is a multi-task AI-text detector using auxiliary heads, uncertainty-weighted losses, EMA distillation, and pairwise ranking that reaches 99.9% TPR at 1% FPR on a new held-out benchmark while remaining competitive on the RAID leaderboard.
-
Block-wise Codeword Embedding for Reliable Multi-bit Text Watermarking
BREW achieves TPR of 0.965 and FPR of 0.02 under 10% synonym substitution by shifting from ECC decoding to designated verification with block voting and local validation.
-
DSIPA: Detecting LLM-Generated Texts via Sentiment-Invariant Patterns Divergence Analysis
DSIPA is a zero-shot black-box detector that uses sentiment distribution consistency and preservation metrics to identify LLM text, reporting up to 49.89% F1 gains over baselines across domains and models.
-
Zero-Shot Detection of LLM-Generated Text via Implicit Reward Model
IRM derives implicit reward signals from off-the-shelf LLMs to detect generated text zero-shot and reports better results than prior zero-shot and supervised detectors on the DetectRL benchmark.
-
Whose Story Gets Told? Positionality and Bias in LLM Summaries of Life Narratives
A proposed pipeline shows LLMs introduce detectable race and gender biases when summarizing life narratives, creating potential for representational harm in research.
-
Towards Real-World Validity in Generative AI Benchmarks: Understanding and Designing Domain-Centered Evaluations for Journalism Practitioners
A human-centered design workshop with journalism practitioners yields an evaluation cookbook and design requirements for contextualized, value-aligned generative AI benchmarks.
-
GigaCheck: Detecting LLM-generated Content via Object-Centric Span Localization
GigaCheck detects LLM-generated text at both document and span levels by combining fine-tuned language-model embeddings with a DETR-like architecture that treats generated intervals as detectable objects.
-
Can AI-Generated Text be Reliably Detected?
Recursive paraphrasing attacks substantially lower detection rates for multiple AI text detectors with only minor quality loss, while a theoretical analysis ties best-case AUROC to total variation distance between human and AI distributions.
-
Ethical and social risks of harm from Language Models
The authors provide a detailed taxonomy of 21 risks associated with language models, covering discrimination, information leaks, misinformation, malicious applications, interaction harms, and societal impacts like job loss and environmental costs.
-
CodeXGLUE: A Machine Learning Benchmark Dataset for Code Understanding and Generation
CodeXGLUE supplies a standardized collection of 10 code-related tasks, 14 datasets, an evaluation platform, and BERT-, GPT-, and encoder-decoder-style baselines.
-
Hidden Human-Like Nature of Machine-Generated Texts: Theory and Detection Enhancement
Reveals hidden human-like spans in machine-generated texts that raise detection complexity and proposes a stacked enhancement framework with hard-EM optimization to improve detectors across LLMs.
-
Multi-Level Contextual Token Relation Modeling for Machine-Generated Text Detection
A multi-level framework that models local and global relations among token detection scores to improve machine-generated text detection with low overhead.
-
DetectRL-X: Towards Reliable Multilingual and Real-World LLM-Generated Text Detection
DetectRL-X is a multilingual benchmark evaluating LLM text detectors on 8 languages, 6 domains, 4 commercial generators, and paraphrasing/perturbation attacks.
-
Alignment Imprint: Zero-Shot AI-Generated Text Detection via Provable Preference Discrepancy
LAPD, derived from the provable preference discrepancy in aligned LLMs, improves zero-shot AI text detection by 45.82% over baselines with claimed statistical dominance over Fast-DetectGPT.
-
Interpretable Stylistic Variation in Human and LLM Writing Across Genres, Models, and Decoding Strategies
Genre and model exert stronger influence on writing style than human/LLM source or decoding strategy in a broad comparison of lexicogrammatical features.
-
Rate-Distortion Optimization for Transformer Inference
A rate-distortion framework for lossy compression of transformer representations yields substantial bitrate savings on language tasks while preserving accuracy, with observed rates aligning to derived information-theoretic bounds.
-
Detecting LLM-Assisted Academic Dishonesty using Keystroke Dynamics
Keystroke dynamics models outperform text-only detectors for spotting LLM-assisted academic dishonesty in practical scenarios, though performance drops under adversarial conditions.
-
Lightweight Stylistic Consistency Profiling: Robust Detection of LLM-Generated Textual Content for Multimedia Moderation
LiSCP detects LLM-generated text via stylistic consistency profiling across paraphrased variants and reports up to 11.79% better cross-domain accuracy plus robustness to adversarial attacks.
-
LLMSniffer: Detecting LLM-Generated Code via GraphCodeBERT and Supervised Contrastive Learning
LLMSniffer improves detection of LLM-generated code on GPTSniffer and Whodunit benchmarks by fine-tuning GraphCodeBERT via two-stage supervised contrastive learning plus preprocessing and MLP classification.
-
An Overview of Catastrophic AI Risks
The paper categorizes sources of catastrophic AI risks into malicious use, AI race, organizational risks, and rogue AIs, providing illustrative stories and mitigation suggestions for each.
- Luminol-AIDetect: Fast Zero-shot Machine-Generated Text Detection based on Perplexity under Text Shuffling