Unverified50% confidenceEventExact time
In 2025 Anthropic published a study that looked deep inside Claude to observe its "thinking" process and found that before writing the second line of a poem, the model had already settled on the rhyming word (such as "rabbit") and then wrote toward that goal.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/19/2026
First Seen
Valid until: 1/17/2027
Sources
Related Entities
Related Claims
UnverifiedAnthropic's official glossary states that the model behind Claude is trained to predict the next word based on the preceding text.71% similarUnverifiedAnthropic的Claude系列在扩展思考模式下会展示部分思考内容，但同样保留了对底层推理的一定程度过滤和摘要67% similarUnverifiedAnthropic has previously used the Pokémon series as a reference for measuring Claude's long-horizon capabilities.67% similarUnverifiedAnthropic发布过关于模型内部表征的可解释性研究（'用显微镜观察Claude'系列研究）67% similarUnverifiedAnthropic的Claude使用自研分词器66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1069868API
curl https://kongchang.com/api/v1/knowledge/claims/1069868MCP
get_claim(id=1069868)