PRISMΔDB
SAE MODEL #2

FEATURE 404

/ Interactions and Mechanisms
Refresh Stats

This page shows the detailed analysis of a specific feature in the Sparse Autoencoder (SAE). It includes the semantic interpretation generated by the LLM, the top activating documents that trigger this feature, and statistical metrics like density and activation distribution.

Semantic Interpretation
gemma3:12b #5

The positive examples consistently describe complex systems and processes involving interactions between different components or mechanisms. This includes microbial interactions, composite material synthesis, parental decision-making, device physics, biosensor development, and organic molecule synthesis. A key feature is the focus on *how* these components interact and the *mechanisms* driving the observed behavior. The high activation suggests the neuron is sensitive to descriptions of interconnectedness, process pathways, and the underlying principles governing these systems. The negative examples, conversely, deal with more isolated procedures, statistical analyses, or clinical observations, lacking the same emphasis on mechanistic interactions. Therefore, the neuron likely activates when texts describe complex interactions and underlying mechanisms.

STATISTICS & DISTRIBUTION
Statistics Explained

Density
Fraction of documents where this feature activates at least once.
Higher density = feature appears frequently across the dataset.

Peak Activation
Maximum activation value observed for this feature over all documents.

Activation Histogram
Distribution of all activation values for this feature. Each bar represents a bin (range) of values, and its height shows how many documents fall in that range.

Density
0.01600
Peak Act
3.76
0.0 Max
Global Context
TOP ACTIVATING CONTEXTS
DOC #673 ANALYZE
ACT: 3.7557
Microbial interactions in the anaerobic oxidation of methane: model simulations constrained by process rates and activity patterns. Proposed syntrophic interactions between the archaeal and bacterial cells mediating anaerobic oxidation of methane coupled with sulfate reduction include electron transfer through (1) the exchange of H2 or small organ…
DOC #527 ANALYZE
ACT: 2.9410
Synthesis and characterization of nano-HA/PA66 composites. Based on the bioactivity and biocompatibility of hydroxyapatite (HA) and the excellent mechanical performance of polyamide 66 (PA66), a composite of nanograde HA with PA66 was designed and fabricated to mimic the structure of biological bone which exhibits a composite of nanograde apatite …
DOC #635 ANALYZE
ACT: 2.9311
Better Parenting through Biomedical Modification: A Case for Pluralism, Deference, and Charity. In this paper, I will argue that the moral assessment of the prenatal selection and postnatal modification biotechnologies requires a nuanced approach, which pays close attention to the variety of sometimes conflicting parental roles and reasons involve…
DOC #100 ANALYZE
ACT: 2.7628
Multi-terminal transport measurements of MoS2 using a van der Waals heterostructure device platform. Atomically thin two-dimensional semiconductors such as MoS2 hold great promise for electrical, optical and mechanical devices and display novel physical phenomena. However, the electron mobility of mono- and few-layer MoS2 has so far been substanti…
DOC #618 ANALYZE
ACT: 2.4654
Multisegment nanowire/nanoparticle hybrid arrays as electrochemical biosensors for simultaneous detection of antibiotics. Antibiotics such as penicillin and tetracycline drugs are widely used in food animals to treat, control, and prevent diseases, and penicillin is approved for use to improve growth rates in pigs and poultry. However, due to the …
DOC #52 ANALYZE
ACT: 2.1740
Stable 5,5'-Substituted 2,2'-Bipyrroles: Building Blocks for Macrocyclic and Materials Chemistry. The preparation and characterization of a family of stable 2,2'-bipyrroles substituted at positions 5 and 5' with thienyl, phenyl, TMS-ethynyl, and vinyl groups is reported herein. The synthesis of these new bipyrroles comprises three steps: formation…
CORRELATIONS
W – Weight-space · similarity between decoder vectors (features that point in similar directions in the embedding space).
D – Data / co-activation · features that tend to fire together on the same documents (co-occurrence in the dataset).