A new AI model, Gemma 4 31B "scotoma-2", has been developed to address common "writing tics" found in AI-generated text, such as excessive adjective use or repetitive phrasing. This model, based on Google's Gemma 4 31B, was fine-tuned using a three-round Direct Preference Optimization process with over 9,300 data pairs to reduce these stylistic quirks. Despite its modifications, scotoma-2 is not uncensored and maintains responsible AI safety features, focusing on improving creative writing and roleplay outputs by making them less predictable. AI
IMPACT This model's focus on reducing repetitive AI writing patterns could lead to more natural and engaging AI-generated content for creative applications.
RANK_REASON The item describes a fine-tuned version of an existing model with specific improvements, presented as a research artifact rather than a commercial product release. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
