Goto

Collaborating Authors

 Large Language Model


DARG: Dynamic Evaluation of Large Language Models via Adaptive Reasoning Graph

Neural Information Processing Systems

The current paradigm of evaluating Large Language Models (LLMs) through static benchmarks comes with significant limitations, such as vulnerability to data contamination and a lack of adaptability to the evolving capabilities of LLMs.


ChatGPT gets 'Lockdown Mode' mode for extra security and privacy

PCWorld

PCWorld reports that OpenAI is launching new security features for ChatGPT, including Lockdown Mode and Elevated Risk labels to combat growing threats. Lockdown Mode restricts external interactions and disables web browsing for high-privacy users, while risk labels clearly mark potentially dangerous features. These updates specifically address prompt injection attacks where malicious prompts attempt to trick the AI into performing harmful actions. OpenAI is launching two new security features in ChatGPT to address growing threats to its AI systems, according to a recent blog post . As AI services increasingly connect to wider parts of the web and more external apps, the risk of so-called "prompt injection attacks" also increases.







SpatialRGPT: Grounded Spatial Reasoning in Vision-Language Models

Neural Information Processing Systems

Vision Language Models (VLMs) have demonstrated remarkable performance in 2D vision and language tasks. However, their ability to reason about spatial arrangements remains limited. In this work, we introduce Spatial Region GPT (SpatialRGPT) to enhance VLMs' spatial perception and reasoning capabilities.