Unverified50% confidenceFactExact time
Since 2023, multiple studies have shown that large language models have approached human-level performance on first-order false-belief reasoning, but their performance on emotional state inference and social norm expectation tasks is uneven.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/11/2026
First Seen
Valid until: 12/10/2026
Sources
Related Entities
Related Claims
Unverified魏杰森（Jason Wei）等人2022年的研究发现，当模型规模超过某个临界点，情感推理等任务的表现会发生非线性跃升（涌现能力）72% similarVerifiedAfter training on massive conversational datasets, modern large language models have acquired a certain degree of emotional reasoning capability.71% similarUnverifiedLarge language models do not truly understand emotions, but rather identify typical patterns of a writing style and perform pattern matching.68% similarUnverifiedA 2023 study published in Nature Human Behaviour showed that ChatGPT exhibits group polarization trends similar to those of humans when simulating group discussions, and that the degree of polarization is positively correlated with the strength of group identity in the prompt.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/894710API
curl https://kongchang.com/api/v1/knowledge/claims/894710MCP
get_claim(id=894710)