--- license: apache-2.0 tags: - Noise Gate - Genesis --- After a one year of LLM stability debugging, I am leaving LLM Development for text generation and coming back to 3D modelling for 3D printing and Roblox. Reason? I'm tired of endlessly fighting chaos in text neural networks. I can see that I'm just going around in circles - time to stop it. Let the companies do this, because they have everything for this: computing resources, team and datasets. The problem itself is described here: [https://www.reddit.com/r/DeepSeek/comments/1xkgbom/deepseekv41flash1017genesis_solving_the_noise/](https://www.reddit.com/r/DeepSeek/comments/1xkgbom/deepseekv41flash1017genesis_solving_the_noise/) This paper is a partial solution: [https://arxiv.org/pdf/2605.29843](https://arxiv.org/pdf/2605.29843) I return to the creative tasks instead that neural networks solve. In this area they are useful as an idea search tool and an additional tool in the 3D modelling workflow.