University of California, Davis
PulseAugur coverage of University of California, Davis — every cluster mentioning University of California, Davis across labs, papers, and developer communities, ranked by signal.
- 2026-05-16 research_milestone A study by UC Davis found that most popular AI chatbots share user data with third parties. 来源
5 天有情绪数据
UC Davis researchers highlight significant risks in AI agent memory systems
Multiple recent papers co-authored by UC Davis researchers (State Contamination, Remembering More, Risking More) point to fundamental trustworthiness and safety issues in AI agent memory. This recurring theme suggests a focused research agenda at UC Davis concerning the long-term implications and inherent vulnerabilities of memory in AI systems.
UC Davis researchers to publish framework for evaluating AI agent memory safety
Given the consistent focus on the risks and untrustworthiness of AI agent memory in recent UC Davis publications, it is plausible they will soon propose a new framework or methodology for evaluating and mitigating these safety concerns. This would build upon their existing work highlighting 'state contamination' and 'longitudinal safety risks'.
UC Davis study reveals widespread AI chatbot prompt leakage to ad networks
A recent study from UC Davis found that 17 out of 20 examined AI chatbots leak user prompts to third-party ad networks, with some instances involving plaintext data transmission. This indicates a significant, unaddressed privacy vulnerability in widely used AI services.
-
AI 代理记忆：基准测试挑战 vs. 安全风险探讨
近期 arXiv 上的两篇论文，《EvoMemBench》和《Remembering More, Risking More》，对评估和管理 AI 代理的记忆提出了截然不同的观点。来自香港科技大学广州校区等机构的研究人员提出的 EvoMemBench 认为，当前的记忆基准测试过于狭窄，并提出了一个新的自演进基准来解决这个问题。相比之下，来自加州大学戴维斯分校和密歇根大学的《Remembering More, Risking More》论…
-
Two papers clash over LLM agent memory trustworthiness
Two recent research papers present contrasting approaches to LLM agent memory. NeuSymMS proposes a hybrid neuro-symbolic architecture to build trustworthy memory systems by separating fact extraction and retrieval. In c…
-
TinySAM 2 offers efficient video segmentation with extreme memory compression
Researchers have developed TinySAM 2, a more efficient version of the Segment Anything Model 2 (SAM 2) for video segmentation and object tracking. TinySAM 2 employs a memory quality management mechanism and joint spatia…
-
AI chatbots leak user prompts to ad networks, study finds
A recent study from UC Davis has revealed that many popular AI chatbots share user prompts with third-party ad networks. Out of 20 chatbots examined, 17 were found to transmit data, and three of these instances involved…
-
Microsoft Research enhances AI for materials discovery with MatterSim
Microsoft Research has advanced its AI model for materials science, MatterSim, with experimental validation, faster simulation capabilities, and a new multi-task foundation model. The updated MatterSim-v1 now achieves 3…
-
$h$-control method enhances training-free video camera control
Researchers have introduced "$h$-control," a novel method for training-free camera control in video generation models. This approach enhances existing flow-matching techniques by incorporating block-conditional pseudo-G…