Kuq Uncertainty Eval

kuq-uncertainty-eval

Knowledge of Knowledge: Exploring Known-Unknowns Uncertainty with Large Language Models — Amayuelas et al. (2023) (arXiv:2305.13712, 2023)

What this evaluates

Evaluates a model's ability to distinguish between questions it can answer confidently (known) and those it cannot (unknown). It probes the model's metacognitive uncertainty articulation and calibration under varying prompt conditions.

Datasets

• KUQ — total ?; splits: test (-1)

Metrics

• F1-score (primary) — range: [0, 1]
  • Harmonic mean of precision and recall for the binary classification of known vs. unknown questions. Computed at the classification threshold that minimizes the Equal Error Rate (EER).
• Equal Error Rate (EER) — range: [0, 1]
  • The operating point on the ROC curve where the false positive rate equals the false negative rate. Lower values indicate better uncertainty calibration.

Input / output format

Input: A natural language question presented to the LLM.

Output: The model's generated response, which is subsequently classified as 'Known' or 'Unknown' based on a threshold applied to its uncertainty score or classification output.

Scoring recipe

def compute_metrics(predictions, gold_labels):
    # predictions: continuous uncertainty scores or logits
    # gold_labels: binary (1=Known, 0=Unknown)
    fpr, tpr, thresholds = roc_curve(gold_labels, predictions)
    eer_idx = np.nanargmin(np.abs((1 - tpr) - fpr))
    eer = fpr[eer_idx]
    threshold = thresholds[eer_idx]
    binary_preds = [1 if p >= threshold else 0 for p in predictions]
    tp = sum(1 for p, g in zip(binary_preds, gold_labels) if p == 1 and g == 1)
    fp = sum(1 for p, g in zip(binary_preds, gold_labels) if p == 1 and g == 0)
    fn = sum(1 for p, g in zip(binary_preds, gold_labels) if p == 0 and g == 1)
    precision = tp / (tp + fp) if (tp + fp) > 0 else 0
    recall = tp / (tp + fn) if (tp + fn) > 0 else 0
    f1 = 2 * precision * recall / (precision + recall) if (precision + recall) > 0 else 0
    return {'EER': eer, 'F1-score': f1}

Common pitfalls

• The paper evaluates uncertainty classification via ROC/EER but does not specify the exact prompt format or whether the model outputs a probability or a direct label.
• Fine-tuning on KUQ improves uncertainty detection but trades off with accuracy on known questions, which is often overlooked when reporting only F1/EER.

Evidence (verbatim from paper)

We present the results on the KUQ evaluation set before fine-tuning for GPT-3.5, GPT-4, and the Llama Models. We also present the results after fine-tuning for Llama-7B, Llama-13B, and its derived chat versions. Figure 3 presents the ROC curve (receiver operating characteristic curve), showing the performance of the classification at different classification thresholds. From the plot, we extract the EER and the corresponding F1-score. These results are presented in Table 4.

Citation

@misc{amayuelas2023kuq,
  title={Knowledge of Knowledge: Exploring Known-Unknowns Uncertainty with Large Language Models},
  author={Amayuelas et al. (2023)},
  year={2023},
  note={arXiv:2305.13712}
}

• arXiv: 2305.13712