Search the full wire by company, model, lab, or keyword. Every story we have ever aggregated.
Persona Collapse in multi-agent LLM simulations: agents converge to homogeneous behavior despite distinct profiles; framework measures Coverage, Uniformity, Complexity.
RouteHead method uses query-dependent attention head selection to improve LLM-based document re-ranking.
Layerwise Convergence Fingerprinting: tuning-free runtime defense detecting backdoors, jailbreaks, and prompt injections in LLMs.
With the emergence of large language models (LLMs) and AI agent frameworks, the human-AI co-work paradigm known as Vibe Coding is changing how people code, making it more accessible and productive. In scientific research, where workflows are more complex and the burden of…
FeatEHR-LLM uses LLMs to generate clinically meaningful features from irregular EHR time series while limiting privacy exposure.
LLMs evaluated as semantic evaluators for ASR, achieving 92-94% human agreement compared to traditional WER metrics.
Nemobot is an interactive environment for creating and deploying LLM-powered game agents across multiple game classes using Claude Shannon's taxonomy.
RedirectQA dataset uses Wikipedia redirects to study LLM memorization of factual knowledge via entity surface form variation.
Transient Turn Injection attack exploits stateless moderation in commercial and open-source LLMs via multi-turn adversarial distribution.
Verbal Process Supervision uses structured natural-language critique as training-free inference scaling, improving GPT-5 reasoning on GPQA, AIME, and LiveCodeBench.
Systematic fairness evaluation across six LLMs on intersectional demographic biases using benchmark datasets.
Preregistered study of 7 LLMs finds they resist motivated investor pressure in fraud detection, contrary to prediction, across 3,360 conversations.
When people say AI will speed up drug development or fear that it will bring about mass layoffs, what they have in mind—whether they know it or not—are AI agents. ChatGPT made large language models a mass consumer product. But to change the world, AI needs to do more than just…
When ChatGPT was released to the public in late 2022, it opened people’s eyes to how easily generative AI could churn out vast amounts of human-seeming text from simple prompts. This quickly caught the attention of criminals, who soon began using large language models to…
SafetyALFRED benchmark evaluates multimodal LLMs (Qwen, Gemma, Gemini) on hazard recognition and mitigation in embodied kitchen environments.
Comparative study of output consistency across GPT-4.1, Claude Sonnet 4.6, Gemini 2.5 Flash for exercise prescriptions with safety analysis.
Empirical study: LLMs measurably alter peer review opinions on clarity, originality at top AI conferences.
Membership inference attacks benchmark against SOTA LLMs to detect training data contamination via black-box methods.
GRASPrune jointly prunes FFN channels and KV heads in LLMs under budget constraints via gated structured pruning.
EDGE-EVAL framework assesses LLM deployment viability via energy, latency, cost constraints beyond accuracy metrics.
RDP LoRA uses geometry-driven trajectory analysis to identify optimal LoRA adapter placement in LLM fine-tuning without training.
LLMs exhibit normative conformity in multi-agent discussions distinct from informational conformity, mechanistically analyzed via new tasks.
DESPITE benchmark (12K tasks) reveals LLMs as robotic planners generate dangerous plans 28.3% of the time despite strong planning ability; scale ≠ safety.
Empirical study comparing auxiliary loss interactions for training conditional depth routing gates in large language models.
CognitiveBench: new benchmark measuring LLM performance on four cognitive dimensions (emotion, stance, thinking style, intention) jointly.
Benchmark of 10 frontier LLM agents on NYU CTF Bench offensive cybersecurity tasks across 7 providers, 200 challenges.
Adaptive Contrastive Decoding mitigates over-refusal in safety-aligned LLMs by keeping non-refusal tokens in candidate lists while maintaining high refusal for malicious queries.
Knowledge Graph-LLM system for manufacturing ML explainability: domain data + ML results stored in structured KG with selective LLM-based retrieval.
RISE: scalable data attribution method for LLMs focusing on output-layer influence hotspots via sketching instead of full gradient indexing.
JumpLoRA uses JumpReLU gating to induce sparsity in LoRA adapters, enabling dynamic parameter isolation for continual learning without catastrophic forgetting.