lowercase p starting a word
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. 27 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 to the letter p when it begins a word, such as pumpkin, pox or peregrine. Turning it up made the model's writing repeat words and drift, so the steering does not show 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.
-
lutein, and zeaxanthin – good for your eyes, skin, immune system, and general health.⏎Consider making your next smoothie a “pumpkin pie” smoothie! Place the following in a blender:⏎•
-
arm around Little Man." (Mildred D. Taylor, _Roll of Thunder, Hear My Cry_)⏎- Are you a word wiz?⏎The "pamper"
-
- Hernias: Hernia is the popular term for a series of different diagnoses, all identified by outpouchings on the ventral aspect of the abdominal wall of pigs. Hernias
-
pox is Kyuta and thicken chicken pox.⏎How to make chickenpox⏎What is chicken pox⏎Chicken Pox in Hindi means chickenpox, which is an infectious disease spread by a virus called varicella z
-
12 import random verb = ['erratic', 'unbelievable', 'unfortunate', 'inconsequential'] fruit = ['raspberry', 'pomegranite', 'bananana', 'grape', 'loganberry
-
to Norway and Sweden, although the Lokke-name has been abandoned and the plants are called “trollhavre” (troll oat) and “pukhagre”.⏎On the other hand, when Lokke, on
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: suppose there are two spectrophotroscopically spectrophotosanathos spectrophotosanathos spectrophotosanathos spectrophotosanathos spectrophotosanathos spectroph
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Mode.com. Retrieved 05, 2012, from http://www.studymode.com/essays/The-Struggle-For-Power-In-The-994323.html⏎"The Struggle for
-
2014|⏎|ISBN: ||9781312744967 (DRM-EPUB)|⏎|Read Aloud: ||not allowed|<|endoftext|>In 1
-
diapers to landfills are being used to generate alternative sources of power that run equipment, vehicles and buildings around the world!⏎Confiscated Booze Converted into Bus-Powering Biogas: nearly 200,000 gallons of
-
/essays/The-Struggle-For-Power-In-The-994323.html>.⏎"The Struggle for
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
- L23:1099 excites
- L18:196 suppresses
- L20:648 suppresses
- L19:884 suppresses
- L17:214 excites
- L22:428 excites
- L22:979 excites
- L22:1780 excites
- L2:1554 excites
- L1:6 suppresses
- L21:571 suppresses
- L19:89 excites
- L26:1564 excites
- L20:322 suppresses
- L8:14 excites
- L0:498 excites
- L6:875 suppresses
- L16:353 weakly suppresses
- L24:938 weakly suppresses
- L18:569 weakly excites
- L6:12 weakly excites
- L20:2984 weakly excites
- L6:13 weakly excites
- L2:202 weakly excites "of" after "many" or "which"
Feeds later layers
None: this is the last layer.
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
- 躬
- nier
- getField
- 目前已
- тогда
- ฤษ
- Hiện
- 现
- ようです
- 闷
Suppresses
- aths
- ages
- ath
- hone
- rob
- arro
- ython
- ará
- rec
- oints