ME News

ME AI news, Thousand Token Wood v2 utilizes four different labs' small models (gpt-oss-20b, MiniCPM3-4B, Nemotron-Mini-4B, and fine-tuned Qwen 0.5B) to power agents in a financial simulation game. The core finding is that the friction in the heterogeneous service layer is due to vLLM 0.22.1 requiring the CUDA toolkit, not the model itself. With a tolerance JSON parsing layer, adding models takes just one config line. Information isolation ensures that insider flags are not in the prompts, and scanning tests validate no leaks. Memory uses emotional summaries truncated to avoid drowning. The fine-tuned 0.5B model achieves 0% self-execution and 100% valid quotes, with the truth firewall guaranteeing zero leaks. The small models are reliable format generators but not reliable reasoners, which can be compensated through structuring, prompts, and fine-tuning. (Source: ME)

ME AI news, Stability AI has dropped a new open-source image generation model, designed to run efficiently on consumer-grade GPUs with only 8 to 12GB of VRAM. Compared to previous open-source versions, this model has improved photo realism and text rendering, while reducing memory usage and inference time through architecture tweaks and quantization techniques. Stability is positioning this release as a resource for indie creators, game modders, and small studios that can't afford large dedicated systems. The company has provided sample notebooks and a straightforward web UI for local deployment, along with guidance to help users apply content filters to avoid generating harmful or illegal images. This release further enriches the community-driven ecosystem of image generation tools and could intensify debates around copyright, content moderation, and the impact on professional artists. (Source: ME)

ME AI news, a new explainer video dives into the major updates of the ChatGPT memory system. This update changes how the assistant retains user info during chats. It's reported that OpenAI has revamped the memory system, enabling ChatGPT to regularly review past conversations on a fixed schedule, extracting key details (like preferences or ongoing projects) and storing them as long-term memory for reuse in future interactions. This brings ChatGPT closer to its competitors, like Anthropic's Claude, which showcases structured memory layers in its interface, although OpenAI's approach seems to focus on building general user profiles rather than a strict chronological memory stack. The video also discusses how the refreshed memory features align with OpenAI's broader product direction, including use cases for agents and providing more personalized assistance over time, while raising questions about transparency, user control, and the frequency of memory refreshes—details that OpenAI has yet to fully clarify. (Source: ME)