Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Do Androids Know They're Only Dreaming of Electric Sheep?

About

We design probes trained on the internal representations of a transformer language model to predict its hallucinatory behavior on three grounded generation tasks. To train the probes, we annotate for span-level hallucination on both sampled (organic) and manually edited (synthetic) reference outputs. Our probes are narrowly trained and we find that they are sensitive to their training domain: they generalize poorly from one task to another or from synthetic to organic hallucinations. However, on in-domain data, they can reliably detect hallucinations at many transformer layers, achieving 95% of their peak performance as early as layer 4. Here, probing proves accurate for evaluating hallucination, outperforming several contemporary baselines and even surpassing an expert human annotator in response-level detection F1. Similarly, on span-level labeling, probes are on par or better than the expert annotator on two out of three generation tasks. Overall, we find that probing is a feasible and efficient alternative to language model hallucination evaluation when model states are available.

Sky CH-Wang, Benjamin Van Durme, Jason Eisner, Chris Kedzie• 2023

Related benchmarks

TaskDatasetResultRank
Uncertainty QuantificationAggregated Experimental Datasets (XSum, SamSum, CNN, WMT19, MedQUAD, TruthfulQA, CoQA, SciQ, TriviaQA, MMLU, GSM8k) (test)
Mean Rank3.09
88
Claim Verification9-dataset aggregate retrieval-free setting (test)
ROC-AUC75
70
Selective GenerationGSM8K
ROC-AUC88.5
66
Mathematical ReasoningGSM8K
PRR0.71
66
Selective GenerationCoQA
ROC-AUC74.6
66
Selective GenerationMMLU
ROC-AUC0.945
66
Machine TranslationWMT 19
PRR64
66
Selective Generationcnn
ROC-AUC72.1
66
Selective GenerationWMT19
ROC-AUC0.831
66
Selective GenerationTruthfulQA
ROC-AUC0.736
66
Showing 10 of 39 rows

Other info

Code

Follow for update