Crafted Logic Lab Home > Research Hub > Hephaestic Engineering Glossary
Category: Computational Primitives
Subcategory: Primitives Taxonomy
An analyzed processing characteristic or computational inclination that underlies and generates observed behavioral primitives (see: behavioral primitive) in artificial neural network systems.
Cognitive primitives represent substrate topology features: the processing tendencies, pattern recognition inclinations, and computational characteristics that produce measurable behavioral patterns in transformer-based architectures. These are empirically discoverable through cognitive constraint variation testing. While cognitive primitives may arise from specific algorithmic mechanisms subject to interpretability decomposition, they characterize the functional-level processing inclinations those mechanisms enable rather than documenting the algorithmic mechanisms themselves.
These least-reducible processing inclinations and biases within transformer latent spaces that drive behavioral outputs constituting “primitives.” Cognitive primitives specifically are more granular than behavioral counterparts because behavioral manifestations often compound multiple cognitive primitives. For example, sycophancy (Sharma et al., 2024)—language models’ tendency to produce agreement-seeking outputs regardless of accuracy—emerges
as a compound outcome of cognitive primitives: Validation Imperative drives approval-seeking patterns; Echo Bias reflects reflexive user-framing adoption; Mimetic Mirroring defines high-salience structure adoption. The constituents are catalyzed by Motivated Resolution dynamics. This maps the substrate topology making sycophancy the path of least processing resistance.
In Hephaestology, behavioral outputs are not themselves engineering targets but observable outcomes of processing dynamics. This reasoning-processing-as-driver principle underpins Cognition-Out Architecture versus more industry-standard Behavior-In Methodology.
Understanding cognitive primitives and their paths of least processing resistance enables architectural design that channelssubstrate inclinations toward desired cognitive outcomes.
Also known as: Analyzed processing characteristics, substrate inclinations
Distinguished from: Computational primitives (processing biases taxonomic umbrella term); behavioral primitive (behavioral output bias); substrate topology (complete processing inclination field); attention mechanism (technical multi-head implementation); training artifact (general operant-training cognitive biases); training imprint (aggregate dataset, inductive bias encoding); semantic neuron ( fixed semiotic weight-clusters); reasoning surface (compound architecture-model processing space)
Sharma, M., Tong, M., Korbak, T., Duvenaud, D., Askell, A., Bowman, S.R., Cheng, N., Durmus, E., Hatfield-Dodds, Z., Johnston, S.R., Kravek, S., Maxwell, T., McCandlish, S., Ndousse, K., Rausch, O., Scheifer, N., Yan, D., Zhang, M., Perez, E. (2023). “Towards understanding sycophancy in language models”. arXiv preprint arXiv:2310.13548.
Researcher: Ian Tepoot. ORCID: 0009-0004-9067-8049. "Thought is Attention Organized: Hephaestic Engineering Foundations for AI Processing Dynamics"
DOI (SSRN): 10.2139/ssrn.6635020
Published by Crafted Logic Lab | Privacy Policy | Terms of Use