'set' as in 'set up'
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. 25 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 fires on the word 'set' when it means to establish or arrange something, as in 'set up an office' or 'set up a warehouse'. When it was turned down, the model got stuck repeating 'establishment' and 'established', which are close in meaning to the trigger.
Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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 cores back on campus.⏎A new laboratory, the Cal-Dendro Tree Ring Laboratory, has been set up to conduct research on the cores, which are expected to reveal valuable data going
-
later known as the Cape Dutch) did not trek eastward, a small number did.⏎The first colonists, who arrived in 1652 to set
-
warehouse:⏎- Choose the right location to set up your warehouse by considering the transportation costs and areas of highest demand.⏎- Make sure that you have sufficient capital to set up the warehouse.⏎- Have an optimized floor plan to increase storage space, prevent
-
nations that have tense relations to continue some kind of working contact. The U.S. has operated “interest sections” in places like Cuba and Iran.⏎South Korea set up a liaison office in the North Korean city of Kaesong last year after
-
has remained mostly unused for more than 20 years. But it was once used as an office to handle U.S.-North Korean matters. The office was set up after an agreement signed in 1995 that gave Sweden permission to
-
to. The Trustee then holds that property for the trust’s beneficiaries. They also invest the trust assets and administer the trust terms according to the terms created by the Settlor.⏎Trusts all fall into one
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 falliversigoupoupoupoupoupoupoupoupoupoupoupoupoupoupoupoupoupoupoupoupoupoupoupoupoupoupoupoupoupoupoupoupoupoupoupoupoupoupoupoupoupoupoupoup
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
use of epoxy-based resins and may be classified as the thermosetting epoxide polymer. Exteriors that receive this coating are resistant to ultra
-
weeks 5 days ago fairst put some water in a air tight box and put salt seasalt works best and put water in the container then let the salt desolve the more salt the better and put ure bait in a zipper bag and pop
-
million jobs being provided by the manufacturing industry, which employs 12 percent of the nation’s women and girls. That steady hum of progress in Dhaka--despite its unscrupulousness--has helped the very poorest break free of
-
to users worldwide. The company has developed a series of networks since it was founded in 1979. The withdrawal of RBGAN occurred because it was superseded by the newer BGAN technology.⏎BGAN signal acquisition requires line-of-site
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
- L24:1292 strongly suppresses
- L24:1472 strongly excites
- L24:591 strongly suppresses
- L24:2351 strongly excites
- L23:1075 strongly suppresses
- L22:907 strongly suppresses
- L24:881 strongly suppresses
- L24:2511 strongly excites
- L23:1136 excites
- L22:2938 suppresses
- L24:1607 suppresses
- L24:1231 suppresses
- L23:2853 suppresses
- L23:698 suppresses
- L22:1476 excites
- L24:1693 excites
- L22:976 suppresses
- L20:810 suppresses
- L22:1531 suppresses
- L23:2430 weakly excites
- L22:1793 weakly excites
- L24:1276 weakly suppresses
- L23:573 weakly excites
- L24:1820 weakly suppresses
Feeds later layers
- L27:624 strongly suppresses
- L27:1741 strongly suppresses
- L26:2332 suppresses
- L26:2576 weakly suppresses
- L27:1233 weakly suppresses
- L27:1148 weakly suppresses
- L27:2138 weakly suppresses
- L27:2365 weakly suppresses
- L27:1225 weakly suppresses
- L27:2773 weakly suppresses
- L27:1158 weakly suppresses
- L26:2889 weakly suppresses
- L26:323 weakly suppresses
- L27:2507 weakly suppresses
- L26:2118 weakly suppresses
- L27:2569 weakly suppresses
- L26:2144 weakly suppresses
- L26:1296 weakly suppresses word-start 'mult' fragment
- L26:1626 weakly excites the word 'I' as subject
- L27:1093 weakly suppresses
- L26:66 weakly excites
- L27:971 weakly excites
- L26:1253 weakly suppresses
- L26:1798 weakly suppresses
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
- orners
- adamente
- itzer
- ウォー
- ič
- уд
- ARING
- Ames
- amide
- ط
Suppresses
- set
- Set
- Set
- (set
- sets
- set
- SET
- SET
- Sets
- 设置