'of' after a quantity or container noun
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. 16 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 'of' when it follows a amount or piece word, like 'instances of', 'amount of', 'one part of', or 'scrap of'. It is quieter, or pushed negative, on 'per' in rates such as 'deaths per year'. Turning it up made the text wander without a clear theme.
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.
-
a class in Java? Local variables − Variables defined inside methods, constructors or blocks are called local variables. Classes define the behavior for objects of its type. Instances of the class Class represent classes and interfaces in a running Java application. If we do
-
, but not in all.⏎A local conservation law is usually expressed mathematically as a continuity equation, a partial differential equation which gives a relation between the amount of the quantity and the "transport" of that quantity. It states that the amount
-
followed by the object name, and use the keyword new: Create an object called "myObj" and print the value of x: You can create multiple objects of one class: You can also create
-
inators x pollination per pollinator – total pollination⏎Pollinator loss due to human influence could be compensated for by an increase in the number of individuals of remaining species, called numerical compensation. Alternatively species loss could be made up for not
-
ilohertz (kHz) passive integrated transponder (PIT) tag. Twenty subadults of each species were implanted with short-term (3 month) acoustic telemetry tags to examine
-
a dhoti?⏎Dhoti-wearing men and women in⏎southern India frequently accessorize their traditional garment with a spare⏎scrap of cloth, called an angavastram. The garment is designed to be worn
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: when I have a new project, the time I have a new project, the time I have a new project, the time I have a new project, the time I have a new project, the time I have a new project, the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
with blood or tissues of infected animals such as a cough or a bite. However, the CDC says that there is now only an average of seven human plague cases per year and the WHO says the mortality rate is estimated to be between 8-
-
health and safety falls under the enormous umbrella of public health.⏎More proof that public health is awesome:⏎- On average, 42,000 deaths per year are prevented among children who receive recommended childhood vaccines .⏎- Diabetes is the leading
-
affects some 8.1 women per 100,000 each year. However, it disproportionately affects Latinas (11.1 per 100,000) and African American women (10.
-
914-1919 stand at 1,300, which accounts for about 8% of the total population, one man out of every six. This war is held to have ended in 1919
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
- L14:1901 suppresses
- L12:2757 excites "of" after a measure like inches or teaspoon
- L15:1634 excites
- L15:894 excites
- L15:1283 excites
- L12:1869 excites
- L13:527 suppresses
- L15:331 excites
- L10:1235 suppresses
- L14:539 weakly excites
- L14:655 weakly excites
- L15:765 weakly excites
- L10:272 weakly suppresses
- L11:1039 weakly suppresses
- L2:2800 weakly excites
- L11:2273 weakly excites
- L15:2096 weakly excites
- L12:1801 weakly suppresses
- L15:2395 weakly suppresses
- L8:1254 weakly excites
- L14:1639 weakly excites "and" joining two nouns
- L12:1020 weakly suppresses
- L15:614 weakly excites
- L14:245 weakly excites
Feeds later layers
- L18:1046 excites
- L17:1687 excites
- L21:1572 excites
- L19:2291 excites
- L18:1728 excites
- L19:1379 suppresses
- L17:2587 suppresses
- L17:827 weakly excites
- L17:1113 weakly suppresses
- L19:590 weakly excites
- L17:1040 weakly suppresses
- L19:2720 weakly excites
- L20:1807 weakly excites
- L17:2722 weakly excites
- L18:2058 weakly excites
- L17:3070 weakly suppresses
- L20:1497 weakly excites
- L18:1477 weakly suppresses
- L24:575 weakly suppresses
- L17:3003 weakly suppresses
- L20:2161 weakly suppresses
- L19:2062 weakly excites
- L19:759 weakly excites
- L19:2048 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
- ElementType
- _formats
- 藜
- 电价
- infer
- infer
- ItemType
- entityType
- Rivers
- FileType
Suppresses
- 干部职工
- isode
- is
- UAL
- iset
- experiments
- veces
- 个
- 工作会议
- MERCHANTABILITY