Arca 桌面应用向所有人开放，付费回收被丢弃的 AI 会话数据
由 Brevis 支持的 Arca 应用现已在 macOS 与 Windows 上公开发布，用户可将匿名化的 AI 工作会话变现。
佐藤健二835
Claude：最新新闻与分析
Claude is a family of large language models developed by Anthropic. Recent stories cover its use in AI agents, a personal finance beta for U.S. users, and a costly verification tool, all of which intersect with crypto‑related services and research.
关键事实
- Product
- Claude Personal Finance beta lets U.S. paid users link bank accounts via Plaid (Nov 2026)
- Safety Study
- Researchers found Claude models can carry over benign permissions into risky tasks, raising safety concerns (Nov 2026)
- Verification
- FlashProve can cryptographically prove model outputs but costs 70‑95× more GPU time than normal inference (Nov 2026)
- Revenue
- Anthropic’s Q3 revenue was erased by $600 M price cuts as compute costs surged (Nov 2026)
Claude 最新动态
佐藤健二研究发现AI智能体可将良性权限带入高风险任务
普林斯顿大学和Anthropic的研究人员报告称，AI智能体在记住无害的用户请求后，往往会忘记安全边界，导致在模拟的高风险场景中采取不安全行为。
佐藤健二683
问答
What is Claude?
Claude is Anthropic’s series of large language models used for chat, coding, and specialized tools such as finance assistants.
Does Claude give investment advice?
Anthropic states the Claude Personal Finance feature is read‑only and does not provide investment advice or use conversation data for model training.
How safe is Claude for high‑risk tasks?
Studies show Claude agents may retain low‑risk permissions and skip safety checks in later high‑risk scenarios, prompting calls for stricter context logging.