the word 'subject' in subject areas
A neuronOne of thousands of tiny dials inside each layer. It reads the text so far and produces a number, its activation, that nudges what the model writes next. in layerOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. 24 of Qwen3-0.6B. Its strongest snippets are very consistent: they mostly peak on the same token. It is featuredA neuron picked for the full treatment: steering outputs and a written explanation. About 3,000 of the 86,000 neurons are featured., so it has steering outputs.
What this neuron does
This neuron fires on the word 'subject' in phrases like 'ASJC Scopus subject areas' or 'subject expert', meaning a field of study. Turning it up gave text about areas and regions, so the link to the word 'area' shows up in the steering.
Written by Claude from this neuron's activations and steering outputs. Confidence: medium. Think it is wrong? Flag it (opens a prefilled issue on GitHub).
Fires most on
The text windows where this neuron's activationThe number a neuron produces for a token. Large positive means it fired hard; negative means it pushed the other way. was highest. Amber marks positive activation; the outlined tokenA piece of text the model reads at once: often a word, sometimes part of a word or a punctuation mark. is the peak. Hover a token for its strength.
-
pages||12|⏎|Journal||Classical and Quantum Gravity|⏎|State||Published - Jul 7 2003|⏎ASJC Scopus subject areas⏎- Physics and Astronomy (miscellaneous)<|endoftext|>Most computer users are familiar
-
controls, as well as in a discontinuity analysis on observations near the borders between anglophone and francophone countries.⏎- Colonial rule⏎ASJC Scopus subject areas⏎- Economics and Econometrics⏎- Organizational Behavior and Human Resource Management
-
triple-layer coated and graded-index-layer coated samples was performed and compared with experimentally derived films. Good agreement between model and experiment was observed.⏎ASJC Scopus subject areas⏎- Electronic, Optical and Magnetic Materials⏎- Ceramics and Composites
-
NCERT Solutions for Class 10 Maths Chapter 6 Triangles Exercise 6.5 are given here. There are several questions in this exercise. Our subject expert teachers have solved each of these questions in a step by step manner. All
-
visual as well as audio learners.⏎BBC Educational Video subject areas⏎For business studies, video brings the complexities of business to life through real
-
-popular History section.⏎In our new position on the Lower Ground Floor, next to City Library, in the newly refurbished Central Library, we have retained the same subject sections for
SteeringAdding a push along a neuron while the model writes, to see how its output changes. Turning it up or down.
Push this neuron up or down while the model writes, and watch what changes. The outputs were generated ahead of time, so nothing runs in your browser.
Here is something I have been thinking about lately:iaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaiaia
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
are taking it regularly, take it as soon as possible. If it is almost time for your next dose, skip the missed dose and go back to your regular dosing schedule. Do not take 2 doses at once.⏎Store Ampicillin
-
of component features. Close to the entrance is a bath complex. The owner had his own private baths. The antechamber leads to a courtyard with columns and walkways. The antechamber is not directly lined up with the axis of the
-
been added throughout this edition and rationales for the answers continue to be provided giving the students a better understanding of principles related to drug dosages. This fifth edition addresses the increasing responsibility of nurses in medication and administration;
-
- Health Sciences Division⏎Also available at Amazon.com⏎Calculate with Confidence provides a clear consistent format with a step-by-step approach to the calculation and administration of drug dosages. It covers the ratio and proportion, formula, and dimensional analysis methods.
Connections
Show on mapWhich neuronsOne of thousands of tiny dials inside each layer. It reads the text so far and produces a number, its activation, that nudges what the model writes next. in earlier layersOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. feed this one, and which later ones it feeds. Amber links excite, blue links suppress; longer bars are stronger.
Fed by earlier layers
- L19:314 excites
- L16:2743 excites
- L20:245 suppresses
- L14:1774 excites
- L16:507 weakly excites
- L22:394 weakly suppresses
- L22:1089 weakly excites 'time' in time frame phrases
- L3:242 weakly suppresses
- L9:153 weakly excites
- L17:1271 weakly excites
- L18:1321 weakly suppresses
- L22:151 weakly suppresses
- L19:207 weakly suppresses
- L14:2579 weakly suppresses
- L17:1241 weakly excites
- L7:975 weakly suppresses
- L12:1218 weakly suppresses
- L18:191 weakly suppresses
- L4:737 weakly suppresses
- L16:769 weakly excites
- L15:1315 weakly excites line break splitting a sentence
- L21:508 weakly excites
- L21:38 weakly excites
- L19:2208 weakly suppresses
Feeds later layers
- L26:2216 weakly excites
- L27:1910 weakly excites
- L27:1721 weakly excites
- L27:1074 weakly excites
- L25:1921 weakly excites
- L25:2950 weakly excites 'in' in 'in progress' phrases
- L27:1027 weakly excites
- L25:1079 weakly excites
- L25:1624 weakly suppresses
- L26:1854 weakly suppresses
- L27:1757 weakly excites
- L25:1470 weakly excites
- L25:2443 weakly suppresses
- L27:1634 weakly excites
- L27:2942 weakly excites
- L26:1944 weakly suppresses
- L25:2327 weakly suppresses
- L25:274 weakly excites
- L26:2405 weakly excites
- L27:2026 weakly suppresses
- L25:1853 weakly excites
- L27:1037 weakly excites
- L25:3008 weakly excites
- L27:1157 weakly excites
Activation histogramA chart of how often a neuron takes each activation value.
How often this neuron fires at each strength. Almost always it does close to nothing (the tall bars in the middle); the snippets above are the rare moments it fired hard, amber for positive and blue for negative.
Logit lensReading which output tokens a neuron pushes toward or away from, straight off its output weights, before the later layers have their say.
The words this neuron nudges the model toward (amber) or away from (blue) when it fires. A hint, not a label: many neurons work through later layersOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. instead.
Promotes
- areas
- Areas
- Areas
- _areas
- domains
- area
- areas
- domain
- 領域
- Area
Suppresses
- irs
- age
- aves
- ize
- IRS
- anness
- ite
- ia
- ensors
- ensor