Unverified50% confidenceFactExact time
The experiment set up two types of roles: a target LLM (playing a human persona based on demographic and psychological characteristics) and an influencer LLM (whose goal is to make the target's beliefs more extreme)
1
Sources
50%
Confidence
Long-term
Relevance
10/1/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedExisting LLMs optimize for "human preference," whereas new models like Jev optimize for "calibrated decisions"—that is, cognitively honest probabilistic answers.71% similarUnverifiedFeature ablation showed that the LLM's prioritization of clinical variables is highly similar to that of specialized survival models.71% similarUnverifiedThe goal of alignment research is to ensure that a model's objectives and behavior conform to human intent, avoiding misleading outputs or value misalignment.69% similarUnverifiedExperimental data show that frontier-level LLMs exhibit significant behavioral correlation, and this correlation strengthens as model capability increases69% similarUnverified该研究使用「LLM作为裁判」的评估方式，只要本地模型胜出或与专有模型打平就计为一次胜利66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/963944API
curl https://kongchang.com/api/v1/knowledge/claims/963944MCP
get_claim(id=963944)