Log-Likelihood Ratio
A classical quantity in statistical hypothesis testing, equal to the log probability of a token under an AI model minus its log probability under a human model. It measures which distribution a token more closely conforms to.
Timeline (last 90 days)
For each token, the log-likelihood ratio under the two LoRA adapters is computed as local evidence of whether that position was generated or modified by AI.
The classical CUSUM algorithm relies on the log-likelihood ratio, but in practical scenarios the distributions before and after the change are often unknown and only limited samples are available, so the likelihood ratio cannot be computed directly.
All Facts (2)
For each token, the log-likelihood ratio under the two LoRA adapters is computed as local evidence of whether that position was generated or modified by AI.
50%UnverifiedThe classical CUSUM algorithm relies on the log-likelihood ratio, but in practical scenarios the distributions before and after the change are often unknown and only limited samples are available, so the likelihood ratio cannot be computed directly.
50%