A developer used Gemini API to label 4,290 Reddit comments about chef's knives for $9, then fine-tuned the open-source GLiNER model on those labels to achieve 0.83 F1 score, replacing expensive API calls with a local model that pays for itself within thousands of comments.
A developer discusses using regex for named entity recognition and categorization tasks. They demonstrate how regex patterns can efficiently handle tasks like date validation, email detection, and money recognition, comparing favorably to neural network approaches while using far fewer computational resources. The post showcases regex expressivity through composable patterns and finite state machines.