Paech et al.

Researchers trace and reduce "slop" style in LLMs

Researchers have identified that the recognizable "slop" or stylistic quirks in large language models, such as the use of em dashes and buzzwords, originate from specific training stages. Analysis of models like Qwen3-4…