Perpetual Models: A Systems Theory of Continuous Input, Endogenous Feedback, and Non-Terminating Generation
Year: 2026
Type: preprint
Abstract: We study a class of systems, termed perpetual models, that continuously receive multichannel input, form feedback from their own output, and continue internal generation after a dialogue turn ends. We distinguish continuously available input, ongoing internal generation, and user-visible expression, and define active and sleep modes, native control signals, and an output sidecar whose interpretations do not return to the core. In a candidate instantiation, new user interaction reinjects prior dialogue and reconstructs the active working state. Separate reads from each source and bounded writes form the complete input. Sleep disables focus and body calls, permits limited dream retention, and activates random perturbations only after a run of IDLE reaches a threshold. We derive local state-norm bounds, restrictions on calls during sleep, direct-input dependency properties, and sidecar independence under fixed inputs. Counterexamples show that these properties do not guarantee semantic responsiveness or long-term effectiveness. We then specify encoder-side carriers, Engram boundaries, and context-state consistency for DeepSeek-V4.1-Flash. No model implementation or experimental data are provided; learning benefits and deployment efficiency remain open empirical questions.Theoretical preprint. No model implementation, training, or empirical evaluation is reported. The record contains English and Chinese editions of the same work. ChatGPT assisted the writing and preparation; zhanghuatao is the sole author and an independent researcher.中文摘要:本文研究一类持续接收多通道输入、通过自身输出形成反馈,并在对话轮结束后继续内部生成的系统,称为永续模型。我们区分输入持续开放、内部生成持续推进和用户可见表达,提出活动与睡眠两种模式、原生控制信号以及不回送识别结果的输出旁路。在一个候选实例中,新用户交互重新注入既有对话并重建活动工作状态,各来源经分别读取和有界写入形成整体输入;睡眠禁用焦点与机体调用,允许有限梦境留存,并仅在连续 IDLE 达到阈值后激活随机扰动。本文推导局部状态范数界、睡眠调用限制、直接输入依赖及固定输入下的旁路独立性,并以反例说明它们不保证语义响应或长期有效性。最后给出 DeepSeek-V4.1-Flash 上的编码侧载体、Engram 边界和上下文状态一致性设计。本文没有模型实现或实验数据,学习收益和部署效率均为待检验问题。
Author: zhanghuatao
Language: English
Topic: Speech and dialogue systems
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225249762
Cites:
Cited by:
Related to: