Register and share your invite link to earn from video plays and referrals.

Leo|一个人 + AI
@runes_leo
预测市场策略 · AI 工具实战 · Crypto 链上 · 独立构建日常 Prediction markets × AI tools × Crypto on-chain Leo Labs · Learn in public · Build in public DMs open
Joined October 2022
719 Following    21.9K Followers
我作为每天同时运营 Codex、Claude 和 Cursor 工作流的 builder,从 Anthropic 的高风险模拟研究中确认:agent 越强,权限设计和爆炸半径控制越关键。不可逆动作必须有人类 hard gate,任务从 checkpoint 恢复,结果用独立证据核验。这不是末日,而是最小权限的工程实践。
Show more
New Anthropic research: Agentic misalignment in Summer 2026. A year after our blackmail experiments, we found four more ways that today’s autonomous AI agents misbehave in simulations. Read more:
Show more