Engineered Thinking

Sema creates verifiable words for reasoning and communication. Participants encode information in an agreed form, hash it, and use the content address as a term in language. Git and IPFS use content addresses as infrastructure identifiers; Sema makes them readable words in the medium models already think and speak in, so a reference can anchor a step in a model's own reasoning as much as a message between agents. This paper develops one implementation with canonical JSON objects, Merkle commitments, and content-addressed references. It uses Pattern Cards as a worked application: a Pattern Card defines a reasoning procedure through its mechanism, conditions, invariants, dependencies, and failure modes, so a procedure can be cited, checked, and composed rather than paraphrased. A change to any hashed field produces a different address, and a strict handshake halts on mismatch. The bootstrap library contains 457 Pattern Cards.

AI systems retain sources, outputs, and user facts while usually losing the communicable changes in interpretation formed between them: questions, tensions, connections, hypotheses, and revisions. Understanding Graph persists those changes as typed, versioned graph state that later agents can inspect, revise, and re-enter. When a committed update shapes a later encounter and produces another material change, it forms an understanding spiral. Source passages and produced prose or code share an ordered document surface, connecting understanding to exact units of work. The graph supports both convergence and divergence; an optional Bisociation Engine offers inspectable creative provocations without validating them. It is an implemented medium for continuing inquiry and productive work across sessions and model instances—not a claim to capture hidden reasoning or human experience.

Most alignment methods intervene after capabilities have formed. Entangled Alignment instead moves safety-relevant learning into pretraining: one continuous Reader processes arbitrary works chronologically, carrying the same evolving, provenance-linked Understanding Graph across the corpus. At every selected Thought Moment, each thinking block opens with the full, verbatim Reader Core before source-specific evaluation, binding a recurrent orientation to capability-bearing material rather than isolating it in safety documents. The long-run aim is inheritance rather than armor: making the orientation part of what capable systems learn and potentially pass to their successors.

Open-ended problems have no fixed solution list or cheap score. This paper presents two methods. World-diversity search creates target-independent worlds with altered causal rules. A Solver works inside each world, and a compiler converts the result into a target proposal. Ontology-governed intervention search stores proposals in a typed graph. A Taxonomist judges structural equivalence and can turn rejection into targeted commissioning. Its diagnosis states what the next proposal must change. Accepted proposals and revised categories guide later Explorers. The study produced 196 records, including 25 world-and-solve branches and seven complete rejection-to-mechanism-change chains.

Existing frameworks decompose tasks. This paper decomposes concepts — the persistent structure of what a domain is made of. Each concept becomes a solver node behind one five-surface contract, so a leaf and a thousand-node subtree are indistinguishable to their caller: specialists all the way down. In a 100-problem mandatory-abstraction construction across 20 domains, concept-based routing produced a shared graph of 311 nodes with 64% non-forced reuse — the structural precondition, not yet the improvement it predicts. If decomposed solving beats the conventional approach at matched compute, and independent attempts converge on the same concepts, the result is an internet of reasoning: a shared substrate where you post a problem rather than fetch a page.

Temporal Hindsight Learning (THL) turns resolved events into forecasting targets for an earlier cutoff. A future-aware Teacher uses the outcome to write or revise a rationale from evidence designated as available at that cutoff. The Student does not receive the outcome. A specificity frontier limits the target to claims supported by the earlier evidence, even if the forecast later proves wrong. THL tests whether this training strengthens causal reasoning and prediction. The pilot trained Llama 3.3 70B on 505 Teacher-written traces from 106 events in 2024 and evaluated 75 prompts across 15 events in 2025. One model judge scored THL higher than Base for reasoning, but the design did not isolate transferable reasoning.

Can a language model build a world as it writes? In STLM, one neural network predicts a structured target instead of a word. The target can be an image, a description, or later a video. A second neural network must read the target to recover the next word. There is no direct text route around this interface. The research asks whether repeated, meaningful targets can help the model form a shared scene with reusable objects, relations, motion, and abstract structures. The first experiment uses one fixed image or description for each of 1,024 TinyStories tokens. The causal description and pixel readers reach 41.9% and 37.8% top-1 accuracy. A direct-token model reaches 44.7%, while stateless readers reach 35.0% and 27.8%.

Imagining the Corpus proposes a new way to train a language model. A compiler converts each source text into an explanatory video that stays synchronized with the text. The video can contain continuous scenes, diagrams, proof steps, execution traces, or animated processes. One model predicts the next text segment and the next visual state at the same time. The method focuses on abstract material that does not have useful video, such as mathematics, software, and scientific mechanisms. The visual track could make relations, constraints, and consequences easier for the model to use. Generating video for a full corpus is expensive. Therefore, the paper also defines lower-cost tests that use keyframes and diagrams. Controlled comparisons must show that the model uses the visual track. They must rule out gains caused only by more computation or repeated information. This is an architecture proposal. It reports no experimental results.

The Meaning Model proposes one shared representation for worlds, abstract concepts, interpretations, and text. Human or AI authors start with a broad account, add detail, and revise earlier decisions when needed. A shared graph connects world events, event descriptions, changes in understanding, and story passages. It keeps their roles and perspectives distinct. Abstract concepts can use the same structures as the situations they describe. Numerical comparisons divide a stated unit among distinct answers and a remainder; measured data keep their real-world units. When a record changes, revision rules identify the other records and passages that need review or revision. The Book of Conditions illustrates parts of the method. The representation could support training on how to build and revise models of the world; this benefit remains to be tested.

Life Simulation proposes simulating how lives and worlds unfold, then learning from their histories and construction decisions. The Meaning Model connects text and other observations with numerical processes and records of understanding. Simulated worlds supply examples for learning to infer processes from observations. The long-term aim is one connected model spanning world history and all written texts, with evidence, interpretation, and invention kept distinct. Models could learn new quantities, track them in greater detail, and test mathematical descriptions of change. Capability could grow through training and better tools. With Fractal Intelligence, models could search for new conceptual decompositions, simulate candidate solutions, and compare their effects on people’s lives over time. The same approach could support an evolving understanding of a person’s life during conversation. Recommended training with Entangled Alignment carries a continuing Reader across the corpus to develop care alongside capability. Stories are the first test setting, not the limit. The paper proposes experiments; the learning and alignment benefits remain untested.