SAE MODEL #2

FEATURE 2260

/ Historical Context & Development

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 #20

The neuron appears to activate when text discusses the historical development of a field, concept, or methodology, particularly when it involves tracing the evolution of ideas or practices over time. This includes discussions of scientific theories (like aphasia models), ecological changes (savannah biome), child development milestones, or athletic performance. The focus is on how something has changed and evolved, often highlighting connections between different periods or individuals. The presence of specific terminology related to historical analysis (e.g., 'palaeobiome', 'furlongstones', 'connecting link') is also a strong indicator. The negative examples lack this historical narrative and focus on technical descriptions or specific methodologies.

STATISTICS & DISTRIBUTION

Density

0.00100

Peak Act

3.59

0.0 Max

Global Context

TOP ACTIVATING CONTEXTS

DOC #620 ANALYZE

ACT: 3.5939

The rise and fall of the Old World savannah fauna and the origins of the African savannah biome. Despite much interest in the ecology and origins of the extensive grassland ecosystems of the modern world, the biogeographic relationships of savannah palaeobiomes of Africa, India and mainland Eurasia have remained unclear. Here we assemble the most …

DOC #658 ANALYZE

ACT: 1.3968

The α₂ adrenergic antagonist fipamezole improves quality of levodopa action in Parkinsonian primates. Reduction in the antiparkinsonian benefit of levodopa is a major complication of long-term levodopa treatment in advanced Parkinson's disease (PD). Such loss of benefit arises because of reduced duration of action and appearance of disabling dyski…

DOC #558 ANALYZE

ACT: 1.3871

Milestones or furlongstones? A study was carried out over five years of 15000 children aged between one day and three months. It was observed that many infants passed their important 'milestones' much earlier than is suggested in present textbooks. A plea is made for discarding the traditional ages for the various landmarks in the development of a…

DOC #808 ANALYZE

ACT: 1.3085

Kurt Goldstein and his nonlocationist thoughts on aphasia-a pioneer of early network theories at the beginning of the twentieth century? In between Carl Wernicke's locationist aphasia concept from 1874 and Norman Geschwind's new connectionist model of human brain functions in 1965, little notice was taken of the historical debate on aphasia and br…

DOC #507 ANALYZE

ACT: 1.2511

Variation in the aerobic demand of running among trained and untrained subjects. Variation in the aerobic demand (VO2) of submaximal running was quantified among trained and untrained subjects stratified by performance capability. Based on a retrospective analysis of seven published studies, maximal aerobic power (VO2max), and submaximal VO2 value…

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).

Ollama Model

Temp

K+

K-

System Prompt

You are a meticulous researcher investigating a specific neuron in a language model. Your task is to determine what behavior this neuron is responsible for: what concepts, topics, or linguistic features does it activate on?

INPUT DESCRIPTION: You will receive two inputs: 1) Maximum Activation Examples and 2) Zero Activation Examples.
1. You will be given several text examples that activate the neuron, along with a number indicating how strongly it was activated. This means there is some feature, concept, or pattern in this text that 'excites' this neuron.
2. You will also be given several text examples that do NOT activate the neuron. This means the feature or concept is not present in these texts.

OUTPUT DESCRIPTION: Given the inputs provided, complete the following tasks.
1. Based on the MAXIMUM ACTIVATION EXAMPLES, list potential topics, concepts, themes, and features they have in common. Be specific. You may need to look at different levels of granularity. List as many as possible. Give greater weight to concepts more prominent in higher-activation examples.
2. Based on the zero activation examples, systematically exclude any topic/concept/feature listed above that also appears in the zero activation examples.
3. Based on the two previous steps, perform a thorough analysis of which feature, concept, or topic, at which level of granularity, is likely to activate this neuron. Use Occam's razor, as long as it fits the evidence provided. Be highly rational and analytical.
4. Based on step 3, summarize this concept in 1-8 words, in the form FINAL: <explanation>. Do NOT return anything after these 1-8 words.

Respond EXCLUSIVELY with valid JSON: {'label': '...', 'description': '...'}

FEATURE 2260

Semantic Interpretation

Statistics Explained

Global Context