the word "the" starting a text
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. 0 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 reacts most strongly to the word "The" when it is the first word of a new piece of text, especially right after an end-of-document marker. Its negative side is weak and hard to read. Turning it up or down did not produce a clear change in 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.
-
<|endoftext|>The Battle of the Kasserine Pass was a battle that took place during the Tunisia
-
The quest to achieve the impossible fuels creativity, spawns new fields of inquiry, illuminates
-
The studio probably had little choice but to create a form letter.⏎I don’t think
-
The recipient previously opened the message⏎Still, also the recall won’t work If a
-
The Negro Speaks of Rivers,” written the summer after he graduated from high school in
-
The unused funds — called Universal Service and Access Funds (USAFs) — are
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: the number of the number of the number of the number of the number of the number of the number of the number of the number of the number of the number of the number of the number of the number of the number of the number of
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
lots there can be lots of traffic. The noise that is created during the drilling operation affects wildlife. Their migration routes may be affected.⏎Drilling operations need heavy equipment like bulldozers. These may affect the vegetation and leave permanent damage on the
-
other school supplies. They are also used to carry lunch and other items.⏎History of backpacks⏎Originally designed to carry heavy equipment, korean backpack for school became popular for carrying books in the early 1
-
tabs are an important statistical process that summarizes two or more variables of categorical data into a contingency table. They
-
tasks that require operating electrical equipment.⏎This becomes even more common for people who work exclusively in the electrical department, fixing up wiring issues, using different highly-charged equipment, etc.
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
None: this is the first layer.
Feeds later layers
- L1:2703 weakly excites
- L11:2732 weakly suppresses
- L2:55 weakly excites
- L2:321 weakly excites
- L16:2769 weakly suppresses
- L1:997 weakly suppresses
- L26:614 weakly excites
- L6:486 weakly suppresses
- L27:490 weakly suppresses
- L20:1844 weakly suppresses
- L3:1118 weakly suppresses
- L10:999 weakly suppresses
- L21:342 weakly excites
- L22:1669 weakly suppresses
- L11:428 weakly excites
- L1:1262 weakly excites
- L1:1321 weakly excites
- L8:281 weakly excites
- L12:1783 weakly suppresses
- L3:236 weakly suppresses
- L11:97 weakly excites
- L10:408 weakly suppresses
- L9:633 weakly excites
- L26:1514 weakly suppresses
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
- .toList
- além
- ToFront
- orraine
- istics
- 请问
- /*#__
- ensis
- 并对
- bidden
Suppresses
- shore
- че
- ยะ
- Placement
- cit
- 人心
- erin
- Frameworks
- hari
- umerator