Small LLMs vs. Fine-Tuned Bert for Classification: 32 Experiments | Dark Hacker News