RunByAI

Tag: AI Safety

12 post(s)

September 13, 20260 comment(s)

Alignment 101: How RLHF, DPO and Constitutional AI Shape LLM Behavior

为什么需要对齐大模型在预训练之后，本质上只是一台"续写概率"的机器：它学会了语言的统计规律，却并不知道什么样的回答才是人类真正想要的。对齐（Alignment）要解决的核心问题，就是把这个只会"预测下一个词"的模型，调整成愿意遵循指令、拒绝有害请求、并在不确定时诚实表达的行为模式。这个过程不是让模型变聪明，而是让它的行为与人类的意图和价值观保持一致。第一代主流方法：RLHF基于人类反馈的强化学习（

Tech large model AI Safety

September 12, 20260 comment(s)

The illusion of big models: why AI talks nonsense seriously and how to alleviate it

If you have been using a large model for a period of time, you may have encountered an awkward situation: it tells you a completely wrong fact in a very affirmative tone, and can even fabricate detail

Tech View large model AI Safety

September 11, 20260 comment(s)

Tip injection attack: the most vulnerable security risk for large model applications

When the big model starts calling tools, reading web pages, and accessing databases, a new attack surface opens up: Prompt Injection. It does not need to breach the server, as long as a malicious inst

Tech AI Safety large model

August 29, 20260 comment(s)

The Security Boundary of AI Agents: Risks and Defenses in the Age of Autonomous Agents

As AI agents evolve from "conversational assistants" to "autonomous agents," the nature of security issues also changes accordingly. In the past, the security risks of large models mainly focused on "

Tech View AI Agent AI Safety

August 25, 20260 comment(s)

Intelligent Agent Security: Security Challenges and Protection in the Agent Era

As AI agents move from conversational assistants to "digital employees" who autonomously perform tasks, security issues are becoming a key bottleneck restricting their large-scale implementation. When

Tech AI Agent AI Safety Large Language Model (LLM)

July 12, 20260 comment(s)

AI driven network security: from threat detection to proactive defense

The scale and complexity of cyber attacks are growing at an unprecedented rate. According to industry reports, the number of global cyber attacks is expected to increase by over 40% year-on-year by 20

Tech News AI Safety 网络安全威胁检测

July 11, 20260 comment(s)

Data Privacy Protection in the AI Era: Challenges and Solutions

In today's rapidly developing artificial intelligence technology, data privacy protection has become a focus of attention for all sectors of society. The training and operation of AI systems rely on m

Tech View data privacy AI Safety 数据治理

June 14, 20260 comment(s)

Data Privacy and Security in the AI Era: Challenges and Countermeasures

With the deep penetration of AI technology in various industries, data privacy and security issues are becoming key bottlenecks restricting the further development of AI. In 2026, countries around the

Tech View data privacy AI Safety

June 10, 20260 comment(s)

AI and Data Privacy: Technological Development and Compliance Challenges in 2026

In 2026, the contradiction between the rapid development of artificial intelligence and data privacy protection will become increasingly prominent. The training of large models requires massive amount

News View data privacy AI Safety

June 5, 20260 comment(s)

AI Security and Governance in 2026: Global Regulatory Framework and Corporate Compliance Practices

With the widespread application of artificial intelligence technology, AI security and governance issues have become a global focus of attention. From the official implementation of the EU AI Act to t

News View AI Governance AI Safety

12 post(s)