Tag
An open-source local AI dungeon app using Gemma 4 and FLUX for text and image generation, fully private and runs under 8GB RAM.
This paper investigates whether role-playing in LLMs changes only outputs or also internal truth representations, using linear probes. It finds that roleplay shifts outputs more than internal beliefs, while emergent misalignment causes larger shifts in internal representations.
User shares positive experience with Gemma4 QAT model, noting quality improvements and speed gains with MTP, and asks others for their experiences.
A lightweight Python framework for local LLM roleplay using Ollama and Phi-3, featuring context preservation and native streaming to prevent character drift.
Gryphe releases Pantheon-Reasoning-27B, an uncensored dense Qwen 3.6 27B model fine-tuned with reasoning traces for enhanced roleplay and narrative generation. It combines roleplay data with full thinking traces to improve character immersion and narrative planning.