mid-word fragments, bracket output
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 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 on word pieces that start a longer word, such as 'Spe' in 'Speaks' or 'Contr' in 'Contrary'. Turned up, the model wrote long runs of curly or square brackets, and turned down it got stuck on 'exposure'.
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.
-
ancient as the world and older than the flow of human blood in human veins.⏎My soul has grown deep like the rivers.⏎Langston Hughes, “The Negro Speaks of Rivers External.”⏎Hughes’ parents separated shortly after he was born,
-
The Negro Speaks of Rivers,” written the summer after he graduated from high school in Cleveland,
-
a venom that causes an immediate, localized pain. Within minutes, a red, raised spot usually develops, followed a day later by a white, pimple-like pustule that itches like crazy.⏎Because each ant can sting repeatedly — and
-
, often host concerts and other events. During the Christmas holiday season, approximately 100,000 Christmas lights sparkle from<|endoftext|>Disseminated Intravascular Coagulation (DIC) in Dogs⏎Clotting factors, such
-
Thus, the increasing demand for energy and exponential surge in consumption stimulates problems such as an increase in carbon emissions, the greenhouse effect and climate change. Contriving a means of saving energy is therefore crucial. Consequently, many governments and organizations
-
(four men and five postmenopausal women, ages 51-57).⏎"We are the first to document improved aortic hemodynamics in prehypertensive but otherwise healthy middle-aged men and women receiving therapeutic doses of watermelon
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: study study study study study study study study study study study study study study study study study study study study study study study study study study study study study study study study study study study study study study study study study study study study study study study study
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
-2 hours depending on the size of your jar. A full melt pool will reach a depth of 1/2 inch. This will allow the fragrance to disperse more completely into the room. (Remember burn your candles for a maximum of
-
Neem oil. Add up to 5ml of the aromatherapy blend into a foot bath with warm, not hot, water, and agitate to disperse the oils. This bend is beneficial for fungal infestions.⏎Is Tag
-
urry, and the slurry will not be able to adhere tightly to the foamed plastic net. After demolding, it will have no strength and will easily disassemble. Hang up even if it has a certain strength, because the slurry
-
the locations of fractures predominantly on the left side of the posterior crania strongly suggest that these injuries were not accidental. The severity of fractures with large pieces of bone dislodged indicate that these were formed from violent blows in a standardised manner (
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
- L24:2317 excites
- L26:2172 excites
- L25:2371 suppresses
- L19:2046 suppresses
- L25:1058 suppresses
- L24:581 excites
- L23:81 weakly suppresses
- L0:92 weakly suppresses
- L22:1728 weakly excites unclear: code assignments and parentheses
- L22:301 weakly excites
- L26:1026 weakly suppresses
- L9:2088 weakly suppresses
- L26:643 weakly excites
- L24:146 weakly suppresses
- L16:76 weakly excites
- L11:841 weakly suppresses
- L25:62 weakly suppresses
- L1:375 weakly suppresses
- L18:1123 weakly suppresses
- L19:1201 weakly suppresses
- L1:2282 weakly suppresses
- L5:1403 weakly suppresses
- L16:1645 weakly excites
- L5:1157 weakly suppresses
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
- {{
- [[
- {{
- [[
- ((
- {{$
- '{{
- {{$
- ">{{
- {{⏎
Suppresses
- ancode
- ит
- 津
- .Generated
- rä
- \Module
- ridged
- ylan
- meis
- 研究人员