On the IPWT Manuscript and LLM

On the Initial Manuscript of IPWT and the Rewriting Plan #

Proof of Ineffective Input, the writer

Have you ever had that feeling? You painstakingly, with all your heart and soul, feed a “smart” system a pile of “raw materials” that you believe to be logically rigorous and well-evidenced. Then, what it spits out, while superficially polished, always has one or two spots that make you want to grab it by the collar and ask, “Where did all that stuff I gave you go?!”

I’ve recently been deeply immersed in such an “existential crisis”—of course, not the kind of digital siege where every second of thought burns through your wallet, but an academic anxiety triggered by a science fiction writer’s struggle with LLMs while attempting “formalized realism.” Specifically, it concerns my initial manuscript on the Integrated Predictive Workspace Theory (IPWT).

From 58 References to 30+ In-text Citations #

Here’s what happened. To lay a solid theoretical foundation for Chain://, I decided not to be content with vague sci-fi concepts but to truly “formalize” it. This meant I needed a theory of consciousness that was rigorous enough and “verifiable” enough. Thus, IPWT came into being.

In conceiving IPWT, I invested a significant amount of time studying the latest literature in neuroscience, information theory, cognitive science, and philosophy. From Predictive Coding Theory (PCT) to the Free Energy Principle (FEP), from Global Workspace Theory (GWT) to Integrated Information Theory (IIT), and then to cutting-edge information theory tools like Partial Information Decomposition (PID) and Integrated Information Decomposition (ΦID), I scavenged like a hungry digital gleaner in the academic ocean for every brick that could piece together my theoretical edifice. I even delved deeply into the astonishing papers by Dr. Luppi’s team, especially how they revealed the “synergistic workspace” in the human brain through ΦID—this was practically empirical evidence tailor-made for my IPWT!

Ultimately, I compiled a list of 58 references. Each one was carefully screened by me to ensure its close relevance to the core concepts of IPWT—such as Instantaneous Information Integration ( $\Omega_t$ ), Predictive Integrity (PI), Workspace Instances (WSI), and the critical role of Synergistic Information (CI) in consciousness. My intention was for these citations to act like synaptic connections in a neural network, firmly anchoring every argument of IPWT to existing scientific consensus, thereby enhancing the theory’s overall Continuous Information Integration (∫Ω) and making it more irreducible in its logic.

Then, I fed this meticulously prepared “knowledge treasure trove,” along with the core arguments of IPWT, to my LLM assistant. My expectation was that it would act like an efficient academic secretary, embedding these citations appropriately into the text, ensuring every argument was well-supported, and elevating the entire theory’s “information integration” to an unprecedented level.

The result? It did indeed generate a well-structured, fluent initial draft. But when I checked the in-text citations word by word, I discovered a frustrating truth: only just over thirty citations were actually used in the text!

At that moment, I felt like a chef who had meticulously prepared 58 dishes, only for the diners to eat half of them and still praise me for “rich dishes and good taste.” Where did those “ineffective inputs” I provided go? Or were they deemed “redundant information” and ruthlessly pruned in the pursuit of textual Predictive Integrity (PI) and fluency?

IPWT: The Birth of a PoC and its “Ω” #

Of course, I must admit that the current IPWT manuscript is a Proof of Concept (PoC). Its birth was to “formalize” a complex theory of consciousness in a very short time and serve as the backbone for the story of “Chain://”.

This PoC successfully achieved its core objective: it integrates the dynamic drive of PCT/FEP, the architectural platform of WT, and performs a functional reconstruction of IIT’s phenomenological axioms, shifting them from physical causal irreducibility to the logical irreducibility of information flow. We introduced Instantaneous Information Integration ( $\Omega_t$ ) as the theoretical gold standard for conscious integration, and used Predictive Integrity (PI) as its computable proxy. These concepts, though perhaps somewhat rough in the initial manuscript, are sufficient to support the core settings in the story regarding “consciousness on-chain,” “existence with a clear price tag,” and “digital siege.”

However, as a PoC, it inevitably bears the marks of “rushed work.” Some concepts may not be elaborated deeply enough, some logical transitions may seem a bit stiff, and those citations “omitted” by the LLM are precisely a manifestation of this “imperfection.” Its current ∫Ω, while sufficient to get the story running, is still a long way from a truly “highly integrated” theory. It’s like a newly built WSI, capable of basic information integration and prediction, but its internal Synergistic Information (CI) may not be rich enough, and the minimization of Prediction Error ($\epsilon$) still has room for improvement.

Pursuing Higher ∫Ω #

Therefore, I’ve decided that this initial manuscript of IPWT is just the beginning. I’ve opened a new Issue for it on my “task list,” preparing for a thorough rewrite. The goal is to elevate IPWT’s ∫Ω to a higher level, making it not only convincing in the story but also capable of withstanding more rigorous scrutiny at the theoretical level.

First, the citation issue must be resolved. I will personally review each of the 58 references, ensuring that every valuable citation finds its rightful place in the text.

Second, diagrams are essential. A complex theory, if it cannot be aided by intuitive diagrams for understanding, will significantly reduce its information transfer efficiency. The diagrams I envision include:

  • Timeline of Consciousness Science Development: Placing IPWT within the historical river of consciousness science, reviewing the evolution from Descartes to GWT, IIT, PCT/FEP, and pointing out how IPWT performs integration and innovation based on these theories. This will provide readers with a broader situational awareness, understanding IPWT’s historical predictions and future directions.
  • WSI Dynamic Diagram: Demonstrating how WSI dynamically forms, integrates information, performs selective broadcasting, and how its Predictive Integrity (PI) and Information Integration ( $\Omega_t$ ) change under different states of consciousness (e.g., waking, sleep, anesthesia). This helps to concretize the concept that “consciousness is a continuous evolution of encrypted states.”
  • Qualia Functional Label Diagram: Attempting to explain through diagrams how Qualia serve as functional labels for internal system states, and how their quantifiable dimensions (e.g., intensity, clarity) relate to information processing parameters within WSI (e.g., precision weighting of prediction error, complexity of synergistic information). This is undoubtedly a bold formalization attempt at the “hard problem.”

Finally, and most importantly, I will supplement with richer clinical evidence. IPWT emphasizes its potential in explaining mental illnesses and consciousness disorders, such as blindsight, psychedelic states, schizophrenia, and dissociative identity disorder (DID). The current PoC only touches upon these, but in the rewrite, I will delve deeper into the neurocomputational mechanisms of these conditions and elaborate on how IPWT provides a unified explanatory framework. For example, how these complex phenomena can be explained through predictive coding dysregulation, WSI dysfunction, and neural gating mechanism dysregulation.

Of course, this will take time, effort, and a lot of Gas—LLM API Credit. So, please consider the current IPWT version as a PoC, an initial predictive frame. I’ve already opened an Issue on GitHub; feel free to follow the latest updates.

Unless I suddenly disappear, probably.