prism-0aa31956·3 events·first seen Aliases: Prism
Researchers introduce PRISM, a single-layer logistic probe over LLM hidden states that distinguishes physically grounded danger (e.g., unsafe robot actions) from ordinary text-level content danger, showing these are separable signals in representations across multiple model families. PRISM achieves 86.2–87.7% accuracy on SafeAgentBench with substantially lower false-positive rates than same-scale LLM judges, which over-block safe tasks at 24.7–39.0% FPR. The authors also release PhysicalSafetyBench-1K (PSB-1K), a 1,000-pair contrastive benchmark for evaluating physical-risk detection without relying on explicit harm keywords. The work is relevant to safety in embodied AI and agentic systems where linguistic safety filters are insufficient.
Prism is an open-source codebase designed to address engineering bottlenecks in Multimodal Continual Instruction Tuning (MCIT) research. It introduces a plugin registration mechanism that separates algorithmic development from backbone MLLM implementation, allowing new continual learning strategies to be integrated without modifying the underlying model codebase. This design aims to eliminate structural fragmentation across method-specific implementations and enable fair, reproducible comparisons at scale.
OpenAI has launched Prism, a free LaTeX-native workspace designed for researchers that integrates GPT-5.2 directly into the writing and collaboration environment. The product targets academic and scientific workflows, combining document authoring with AI-assisted reasoning in a single interface. This also marks a public reference to GPT-5.2, indicating a model iteration beyond GPT-5.