unclear: mostly line breaks, mixed 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 mixed: they peak on different tokens. 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 weakly and on a mix of things, most often a line break at the end of a line of text. Turning it up made the model write about random topics, such as math or weather, with no clear link to the trigger.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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.
-
People normally<|endoftext|>Spring Dawn by Meng Haoran 春眠不觉晓 处处闻啼茑 夜来风雨声 花落知多少⏎Pin Yin Version Chun mian bu jue xiao Chu chu wen ti niao
-
东方在线 发布人:young⏎14. Energy Resources⏎China is rich in various energy resources. Coal and
-
��性病的⏎anachronism /əˈnækrəˌnɪzə
-
Asia, and America. The American variety (var. Americana) is larger. Called⏎whistler, garrot, gowdy, pied widgeon, whiteside, curre, and doucker. Barrow's golden
-
ized bed boilers; Circulating fluidized bed boilers⏎- Front<|endoftext|>CNY - 中國元,人民幣⏎最常用 CNY 交叉匯率⏎為什麼您
-
, with the introduction of flying boat service to Asia via Pan Am’s famous clippers, Midway became an important refueling and stopover point until war intruded in 1941.⏎Beginning in 1940, facilities
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: 200 The user is a person with a very special way of explaining. The user is a person with a very special way of explaining. I have The user is a person with a very special way of explaining. I
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
“, and others in my awesome Edumatch Voxer group, told me about Break
-
soil should be used for building and development,” Bradley explained. “In the whole talk of food production and global warming, soil fertility gets lost in the discourse.”⏎Assistant Professor of Economics Erik Nelson, Bradley’s advisor, says his student’s work is
-
rights and freedoms,” National Congress of Australia’s First Peoples co-chair Rod Little said.⏎Assistant Minister for Social Services and Multicultural Affairs Minister Zed Seselja described the
-
are never filled. Such shortcomings in medication adherence can decrease the effectiveness of a treatment. This increases the likelihood that
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
- L4:1965 weakly excites
- L9:232 weakly suppresses
- L1:1175 weakly suppresses
- L25:1669 weakly suppresses
- L6:1281 weakly suppresses
- L6:1168 weakly excites
- L17:55 weakly excites
- L1:699 weakly suppresses
- L14:2997 weakly suppresses
- L5:1023 weakly suppresses
- L16:2560 weakly suppresses
- L13:1326 weakly suppresses
- L21:351 weakly suppresses
- L1:856 weakly excites
- L23:1354 weakly suppresses
- L23:512 weakly suppresses
- L13:885 weakly suppresses
- L12:2955 weakly excites
- L17:1676 weakly suppresses
- L18:583 weakly suppresses
- L24:3065 weakly excites
- L19:2191 weakly excites
- L26:1907 weakly suppresses
- L20:2188 weakly suppresses "enough" before "to"
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
- oen
- amb
- 繁
- mins
- min
- .Binding
- 分钟左右
- cos
- mins
- len
Suppresses
- erate
- itore
- 屐
- ycop
- gars
- PointerException
- estroy
- potatoes
- xEE
- orrar