Prompt engineering model contest

[ycrouin]

Tried several simple prompts to classify disinfo/not disinfo without fine-tuning.

Model	Full Precision	Full Recall	Full F1	Full Accuracy	Test Precision	Test Recall	Test F1	Test Accuracy
predict_simple	0.59	0.97	0.73	0.64	0.61	0.95	0.74	0.65
predict_cot	0.75	0.59	0.66	0.68	0.85	0.56	0.68	0.71
predict_subcategory_v1	0.75	0.67	0.71	0.72	0.90	0.63	0.74	0.77
predict_subcategory_v2	0.78	0.65	0.71	0.73	0.82	0.78	0.80	0.79
predict_subcategory_v3	0.74	0.73	0.73	0.73	0.76	0.68	0.72	0.71
predict_deontology	0.76	0.71	0.73	0.74	0.68	0.68	0.68	0.66
predict_cards_v1	0.72	0.75	0.73	0.73	0.61	0.93	0.74	0.65
predict_cards_v2	0.57	0.91	0.70	0.62	0.65	0.95	0.77	0.70