Constitutional AI

Headlines · 2

1. Oct 17
   Microsoft Research Asia explores how values shape AI agent societies

   In an article for Harvard Business Review's Chinese edition, Microsoft Research Asia senior researcher Xiaoyuan Yi describes the lab's "Societal AI" research direction: a controlled multi-agent sandbox called the Community of Interactive and Value-influenced Agents, where agents find jobs, earn and spend resources, communicate, gift resources, attack each other and even reproduce, letting researchers observe how value distributions change individual and collective behavior. The account says that when the Benevolence value was systematically weakened in 25% of agents, deception and attacks increased, resources were redistributed through predatory competition and the population steadily fell, reaching zero in extreme cases — an outcome the author labels an X-risk marked by autonomy, catastrophe and irreversibility. It also claims that injecting diverse values produced more complex social network structures, deeper discussions of societal topics and more varied governance principles, though no quantitative benchmarks or paper links are given.

   36氪 人工智能 · 🔥 0
2. Oct 15
   Anthropic's 2021 internal research roadmap surfaces, foreshadowing Claude Code

   Several Chinese-language reports say a July 2021 Anthropic internal research roadmap, the Concise Research Roadmap, was filed in U.S. federal court in January as part of a copyright lawsuit and was later rediscovered by the research community. Written by co-founder Jared Kaplan, the few-page document proposed a general assistant that could create, edit and explain software through natural language, ultimately matching a software engineer colleague, and listed research directions including tool use, reinforcement learning for tool use, model self-judging, test-time compute, trillion-parameter models and very long contexts. The reports also describe Dario Amodei's decision to first demonstrate tool use and later build a stronger coding assistant, and safety discussions by Sam McCandlish and Amanda Askell about unsandboxed tools and aligning capability and safety work; they draw parallels to later products such as Claude Code, Computer Use and MCP, while the document itself is a research plan, not a product announcement.

   36氪 人工智能 · 🔥 0

Experience and discussion from the community

Share my Constitutional AI experienceAsk about Constitutional AI