Issue 2026-10-19 · Industry · 开源 · 研究
Retrained Qwen3-4B removes AI writing tics

Hacker News discussion covers a version of Qwen3-4B trained to remove AI writing tics. The post compares one sampled answer each from the original and released models on 40 held-out prompts, underlining passages counted by a slop detector.
Read original ↗Topics:Qwen3-4B
- Specializing SLM vocabularies can cut inference cost by 77%
- Intel Launches Inference Microservices for Tuned LLM Inference
- Qwen3-4B post-trained to cut reasoning tokens by 44%
- Framework-free prototype learner Jayce trains 1.6x-4x faster than backprop
- ProgramAsWeights Compiles English into Local Neural Programs