To predict the behavior of people in these documents effectively, representing their emotional states is likely helpful, as predicting what a person will say or do next often requires understanding their emotional state.
情绪表征不是 Anthropic 有意训练的结果，而是预训练阶段的「副产品」：为了预测人类文本中的下一个词，模型被迫学会了理解情绪。令人惊讶的是，这个能力在后训练阶段被「复用」来驱动 AI 助手的行为，形成了一条没有人刻意设计的情绪回路。
But in reality, what we witness is the emergent patterns from each individual starling simply doing it darndest not to collide with the starlings nearby.
Biologists, physicists, ecologists and other scientists use the term “emergent” to describe self-organizing, collective behaviors that appear when a large collection of things acts as one. Combinations of lifeless atoms give rise to living cells; water molecules create waves; murmurations of starlings swoop through the sky in changing but identifiable patterns; cells make muscles move and hearts beat. Critically, emergent abilities show up in systems that involve lots of individual parts. But researchers have only recently been able to document these abilities in LLMs as those models have grown to enormous sizes.
From smaller components, larger and more complex systems are built. In other fields, emergent behavior can be predicted. In LLMs, this ability has been unpredictable as of yet.
An emergent behavior or emergent property can appear when a number of simple entities (agents) operate in an environment, forming more complex behaviors as a collective.
This creates what is essentially an evolution process for the program, causing it to depart from the original engineered design. As a consequence of this and a changing environment, assumptions made by the original designers may be invalidated, introducing bugs.