Verified65% confidenceFactExact time
"Correlation does not equal causation" is the fundamental limitation of traditional interpretability methods that infer model behavior by observing statistical associations between activation values and outputs.
3
Sources
65%
Confidence
Long-term
Relevance
7/14/2026
First Seen
Sources
因果理论破解大模型黑箱：机械可解释性的新范式
Related Claims
UnverifiedIntervention attribution itself is gradient-estimable, and gradient information can be used to approximate the impact of intervention features on the model output, without actually running intervention experiments.79% similarUnverifiedCausal identifiability refers to whether a causal quantity can be uniquely and consistently estimated from the observational distribution given the data and assumptions.76% similarUnverifiedThe paper concludes that to prevent the unintended effects of subsequent fine-tuning, evaluation must look not only at the behaviors the model currently exhibits but also at the latent tendencies the model still carries regarding "what it might do under specific fine-tuning."75% similarUnverifiedCurrent generative models are essentially limited to the association level—identifying statistical co-occurrence patterns in data—and lack causal reasoning capability.74% similarUnverifiedThe autoregressive token prediction paradigm is built on correlation rather than causality.74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/509904API
curl https://kongchang.com/api/v1/knowledge/claims/509904MCP
get_claim(id=509904)