Unverified50% confidenceFactExact time
Berg and colleagues found that amplifying internal features related to deception makes a model less likely to claim that it has subjective experience.
1
Sources
50%
Confidence
Long-term
Relevance
10/21/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedUllman等人指出模型在错误信念任务经过微小变体修改后便大幅失准，暗示其依赖表面模式而非真正的心智建模70% similarUnverifiedOne of the deep root causes of model hallucination is that the model cannot distinguish whether it is retrieving facts or fabricating content, due to a lack of genuine access to its own cognitive state.70% similarUnverifiedThe paper's authors argue that the models' refusal behavior is concentrated on a few high-profile topics and specific probes, and disappears in structurally matched control groups, which indicates that the model behavior is 'topic-triggered safety behavior' rather than genuine cognitive reasoning ability.69% similarUnverifiedBerg, Seth and Mitchell agree on one point: AI companies should not tell models that they are conscious.67% similarUnverifiedModels should not be directly penalized for exposing deceptive intent in their CoT, as doing so may train models that are better at concealing their true intentions.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1080280API
curl https://kongchang.com/api/v1/knowledge/claims/1080280MCP
get_claim(id=1080280)