Safety & Evaluation
A
Showing 1–30 of 71
-
Guarding the Gates of KnowledgeIEEE Spectrum editor on AI-faked papers and publishers' gatekeeping struggleThe author argues AI answers still can't be trusted, so users must verify. Applied at scale, that maxim challenges publishers guarding the scientific record of peer-reviewed work. A feature on faking a paper with AI shows the human gatekeepers trying to hold the line.
-
GhostPrefix: Synchronization-Free Universal Prefix Attacks for ASR Manipulation
-
Aligning MLLMs with the Latent Structure of Human Cognition via Behavior-Derived Semantic Dimensions
-
ReX-GG: a LLM Ensemble Pipeline for Relation-extraction and Graph Generation
-
Iterated Integrals as Lexical Fingerprints: Path Signature Embeddings for Lexical Similarity
-
RECAST: REconstruction of Chain-of-Thought via Auditable Structured Traces
-
Backpropagation-free Deep Learning for Time-Series Classification
-
What Standard Evaluations Hide About Detectors of Hijacked LLM Agents: Rare Attacks, Unseen Tools and Benchmark Transfer
-
Whose Values, Which Contexts, and What Signals? Grounding Cultural Value Alignment with Re3Culture
-
Peeling the Matryoshka: Extracting Ontological Structure from Retrieval Embeddings
-
Clinical-Protocol-Aware Cyber Resilience for Connected Clinical Technologies Under AI-Enabled Adversarial Conditions
-
When Steering Shifts Discourse: Identifying and Intervening on Discourse-Level Changes due to Activation Steering in LLMs
-
KV-Cache Patching: Tracing How and Where Fine-Tuned Behaviors Arise
-
Community safety report and new protections for teensOpenAI expands ChatGPT violence prevention, adds protections for teensOpenAI says it is strengthening violence prevention in ChatGPT with expanded safeguards and earlier interventions, along with a new commitment to public safety reporting. The post's title also points to new protections for teens.
-
A prospective randomized study of patient-centered AI for clinical visit preparation
-
ShallowDefend: Harnessing Shallow Safety Alignment for Multi-Turn Defender
-
How I Aligned Your Model: Scaling Up Multilingual Safety Training with Templates
-
A Low-Rank Space Captures Linear Alignment Between LLM and Human Neural Activity During Naturalistic Language Listening
-
Training language models in simulation for psychiatric intake
-
An updated Model Spec for more capable and beneficial AIOpenAI updates Model Spec with five core values for capable AIOpenAI published an updated Model Spec describing five core values meant to guide more capable AI toward sound judgment, respect for human agency, and faithfulness to human intent.
-
SCARL: Sequence Confidence-Aligned Reinforcement Learning
-
The Athlete as a Scientist of One: Agentic Bayesian Experimental Design for Individualized Physiology from Consumer Wearables
-
Latent Prediction Yields Cross-Subject Alignable Representations in Sensory-Evoked fMRI
-
Learned Terminal Costs for Short-Horizon Model Predictive Control of Autonomous Trucks in Interactive Highway Traffic
-
Simplifying safety to drive savingsSimplifying safety to drive savings: Fastenal's consolidationArticle on how Fastenal used consolidation to standardize data center spend, simplifying safety purchasing to drive savings.
-
An Interview with Palantir CTO Shyam Sankar About the AI Religious WarStratechery interviews Palantir CTO Shyam Sankar on the AI 'religious war'Stratechery interviews Palantir CTO Shyam Sankar about AI diffusion and safety, the role of forward-deployed engineers, and what the battle in tech is actually about, framed in the title as an AI 'religious war'.
-
How we detect frontier model misalignmentOpenAI details layered monitoring for frontier model misalignmentOpenAI describes how it uses layered monitoring to detect misalignment in frontier models and unauthorized agent behavior, so responders can step in before incidents escalate.
-
Safe Multi-Objective Reinforcement Learning: Zero Constraint Violation and Primal-Averaging Convergence
-
More Is Better? Exploring Latent Correspondence in Federated Hypergraph Learning
-
IMRouter: Imbalance-aware Temporal Pattern Routing for LLM-based Time Series Forecasting