"nothing/something to" before 'do' or 'be'
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. 4 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
Fires on the word "to" right after "nothing" or "anything" or "something", as in "nothing to do with" or "anything to be seen." Turning it up made the model write long, rambling text that did not clearly change 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.
-
the earth. They don’t have anything to do with Pilgrims, of course, but were a tradition recounted in many
-
atypical from the others because of that. In a way if you don’t share your memories, people will find you boring and dull, because you have nothing to talk about, because everything you talk about is a memory. A person cannot talk
-
of this humongous ship, but many believe it had something to do with low grade rivets.⏎Wealthier passengers were offered luxurious dining tables
-
ictions like gambling, that have nothing to do with food. Tolerance is both a result and a cause of addictive behavior
-
their inventions: a 3D-printed laundry soap box that keeps an eye on its volume and orders more when it’s running down.⏎Without any power source to worry about, these devices are decidedly eco-friendly, and very cheap to manufacture.⏎
-
Before DNA sequencing, when bacteria were classified morphologically, W. persica and W. melaphagi were also thought to be related, but these have nothing to do with the famous Wolbachia (W. persica is a
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 other one, and I have to be there too. I need to be there too. Are you sure you want to get through the door in this 300 photoshoot of the human being, the other one, and I
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
not the norm,” said Joe Walsh, a member of the Friends of Matthias Baldwin Park, in an email to National Review. “He was BLM [Black Lives Matter] before there was a slogan.”⏎‘The irony of vandalizing a monument
-
Coalition for Western Women’s History decries this ongoing violence against Black women and men and stands in solidarity and support with Black Lives Matter. As scholars of the history of women, gender, and sexuality we want
-
54th Union Army regiment, the first all-volunteer black regiment. The sculpture, dedicated to black soldiers, is located in Boston.⏎The Black Lives Matter protests and violent riots were sparked by the death of George Floyd, a black
-
of neglect they have experienced in America both in life and in death.⏎We also want to acknowledge and amplify the work that Black women activists have done in founding Black Lives Matter (Patrisse Khann-Cullors, Alicia Garza, and
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
- L2:111 weakly suppresses
- L2:126 weakly excites
- L3:456 weakly suppresses
- L3:950 weakly excites
- L0:943 weakly suppresses
- L2:1608 weakly suppresses
- L3:1712 weakly excites
- L2:1935 weakly suppresses
- L3:1698 weakly excites
- L3:1055 weakly excites
- L0:1103 weakly excites
- L3:2398 weakly excites
- L2:2462 weakly excites
- L2:2248 weakly suppresses
- L3:1015 weakly excites
- L0:786 weakly suppresses
- L2:2782 weakly suppresses
- L0:125 weakly suppresses
- L0:3066 weakly excites
- L3:2368 weakly suppresses
- L3:1306 weakly suppresses
- L1:2008 weakly excites
- L3:2879 weakly excites
- L1:1209 weakly excites
Feeds later layers
- L15:203 weakly excites
- L5:2538 weakly excites
- L5:1969 weakly excites
- L5:2289 weakly suppresses
- L10:2155 weakly excites
- L6:2211 weakly excites
- L6:2846 weakly suppresses
- L9:285 weakly excites
- L20:1702 weakly suppresses
- L15:2014 weakly excites
- L13:2296 weakly suppresses
- L17:608 weakly excites
- L22:2405 weakly suppresses
- L25:1699 weakly suppresses
- L8:2640 weakly excites
- L14:938 weakly suppresses
- L20:1956 weakly excites
- L8:2449 weakly suppresses
- L10:3000 weakly suppresses
- L5:1206 weakly suppresses
- L11:666 weakly suppresses
- L5:2742 weakly excites
- L15:1633 weakly suppresses
- L13:844 weakly suppresses 'of' after a percentage or share
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
- enses
- agoon
- 法
- isArray
- AdminController
- 目的
- adoras
- 伊斯
- reason
- NAS
Suppresses
- gnu
- 坟
- bersome
- 绩
- 邮
- ombre
- 等奖
- tire
- tree
- ẩn