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.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Claim Verification | 9-dataset aggregate retrieval-free setting (test) | ROC-AUC75 | 70 | |
| Misclassification Detection | COLA | ROC-AUC61.5 | 31 | |
| Hallucination Detection | CDM (test) | F1 Score75 | 16 | |
| Uncertainty Estimation | Aggregate (Cola, GEmot, IMDB, News, SST5, Toxigen, YELP) | ECE10.3 | 13 | |
| Hallucination Detection | CF (test) | F1 Score94 | 10 | |
| Hallucination Detection | E2E (test) | F1-R90 | 10 | |
| Span-level classification | CDM (test) | F1-Sp0.55 | 6 | |
| Span-level classification | E2E (test) | F1 Score (Span)56 | 6 | |
| Span-level classification | Conv-FEVER (CF) (test) | F1 Score (Spans)0.81 | 6 |