Show HN: Qwen3-4B with the AI writing tics trained out

A developer trained Qwen3-4B to remove common AI writing habits (like overused em dashes, rhetorical questions, and flowery language) and released it as qwen3-4b-deslop, reducing its "slop score" from 121 to 12.5 per 1000 words while maintaining factual accuracy and genre compliance. The model was fine-tuned using distillation techniques on just 161 answers from a MacBook Air, with the resulting 2.5GB model now writing in a plainer style that scores similarly to human writing on 68% of test answers. The approach is model-size agnostic and can theoretically be applied to larger models, though the de-slopped version still has limitations like occasionally inventing names and places.