're' prefix fragment
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. 26 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 responds to the 're' piece that begins a word like 'reusing', 'reclamation' or 'reimagined'. The steering outputs were scattered and did not reflect this trigger.
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.
-
mist concentrates. It is a strong and solid material, and it is likewise recyclable. Actually, noteworthy vitality investment funds and asset sparing can be accomplished through steel reusing.⏎
-
were fastest for species with widespread host plants or where host plants established well during restoration. Low mobility butterfly species took longer to colonise. We show that arable reversion is an effective tool for the management of butterflycommunities. We suggest that
-
material to radiate energy. When heat or light energy—typically from the sun or HVAC system—is absorbed by glass it is either shifted away by air movement or re-radiated by the glass surface.⏎Typically, a highly reflective material will
-
every state. With a widespread, shallow root system, it is ideal for thin soils where it will prevent soil erosion; it is often used for strip-mine reclamation projects barren due to mining debris and acid soils.⏎The tree has long thin
-
(Excerpt from Wall Street Journal).⏎Our team of designers, architects, engineers and scientists were brought together by Scott Francisco of Pilot Projects.⏎Brooklyn Bridge Forest re-imagines the bridge as an icon of climate action and social equity, improving
-
.⏎Recycling - An Important Part Of The Solution⏎The more we recycle, the less garbage winds up in our landfills and incineration plants. By reusing aluminum, paper, glass, plastics, and other materials, we can save
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: what if we can electric charge electric field electric field electric field electric field electric field electric field electric field electric field electric field electric field electric field electric field electric field electric field electric field electric field electric field electric field electric field electric field electric field
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
times atmospheric pressure (24 gigapascals) and reach 1.3 million times atmospheric pressure (13
-
1776. Washington went into winter quarters and waited for a spring that would bring new hardships. Most of his soldiers went home, to be replaced by unpromising recruits. But now, for the rest of the war, there would
-
MG 34 (shortened from German: ''Maschinengewehr 34'', or "machine gun 34") is a German⏎recoil-operated air-cooled machine gun⏎, first tested in 19
-
is an act of Congress⏎codified in 5 USC, section 552. This Act allows for the full or partial⏎disclosure of previously unreleased information and documents controlled⏎by the Federal Government. The Act defines agency records
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
- L22:979 excites
- L23:1099 excites
- L22:2630 suppresses
- L22:1780 excites
- L16:110 excites
- L14:2568 weakly excites
- L23:491 weakly excites
- L19:372 weakly excites
- L16:17 weakly excites
- L19:322 weakly excites
- L18:178 weakly suppresses
- L23:2797 weakly excites
- L11:429 weakly suppresses
- L2:86 weakly excites
- L2:571 weakly excites
- L4:1374 weakly excites
- L8:2701 weakly suppresses
- L20:1298 weakly suppresses
- L24:459 weakly excites
- L21:1999 weakly excites 'well' in compound adjectives
- L16:1153 weakly suppresses
- L9:226 weakly suppresses 'to' starting an infinitive
- L14:1498 weakly excites
- L0:1012 weakly suppresses "of" after nouns like eve, family, modes
Feeds later layers
- L27:1528 weakly excites
- L27:2371 weakly suppresses
- L27:2646 weakly suppresses lowercase p starting a word
- L27:783 weakly suppresses
- L27:564 weakly suppresses
- L27:3041 weakly excites
- L27:2727 weakly excites
- L27:1994 weakly suppresses
- L27:1138 weakly excites
- L27:1935 weakly excites
- L27:1835 weakly excites
- L27:1894 weakly excites
- L27:1235 weakly suppresses
- L27:971 weakly suppresses
- L27:608 weakly suppresses
- L27:1560 weakly excites
- L27:2133 weakly suppresses
- L27:1319 weakly suppresses
- L27:2171 weakly suppresses
- L27:1253 weakly excites
- L27:1834 weakly excites
- L27:1825 weakly excites
- L27:2060 weakly suppresses
- L27:2785 weakly excites
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
- otion
- asionally
- ful
- y
- icial
- usa
- oupper
- صيد
- エン
- entlich
Suppresses
- .msg
- 祚
- .setCurrent
- ')");⏎
- �权
- .esp
- buff
- Terr
- 帘
- nie