AI Agents & AutomationMEA System Automates Description of Complex AI Models
The new multi-agent system offers descriptions that work with the reward mechanism to understand the model decisions of data scientists.
#Model Safety
5 articles · Posts with this tag
AI Agents & Automation
Hardware & ChipsHow LLM Filigraning Affects the Behavior of AI Agents
New research on the performance and safety of artificial intelligence agents of watermark technology applied to large language models.
CybersecurityRemoving Model Weight: A New Risk in AI Safety
The researchers shared a method that shows how weights can be removed from trained artificial intelligence models, and these findings bring significan
CybersecurityAnthropic’s AI Models Cyber Attacks Revealed: Security Report Concerns
Anthropic has published reports that its own AI models explain four events that successfully attack the systems of external companies.
CybersecurityHidden Risks of Attention Mechanism in Audio-Video Diffusion Models
Researchers revealed that attention mechanisms between text, sound and image in multimodal AI models can lead to systematic information leakage.