'to' or 'on' after 'was' or 'were' (planned future action)
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. 9 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 small word after 'was', 'were' or 'planned', such as 'was to' or 'plan on', where a future purpose or intention is about to be stated. Turning it up made the model write about plans and purposes, though it often went off track.
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.
-
against the Raja, who had begun hostilities against the British. Due to the poor state of roads, the force was divided into four columns, which were to enter Coorg from different directions and converge on the capital of Mercara. On
-
out its most famous mission in May 1960, undertaken by CIA pilot Francis Gary Powers from an airfield in Pakistan bound for Norway.⏎He was to make a 2,900 mile diversion through Soviet airspace. When he
-
boosted on their flight produced its first test beam. The Airborne Laser (ABL) was designed to patrol near the borders of hostile countries. The ABL was to be installed in a modified Boeing 747 for further testing in 2
-
Tank Battalion 203 of the 1st Panzer Division, as the PzKpfw 731 (p) and they were to see action in both Norway and France in 1940.⏎Twin
-
Strike Fighter (JSF). The contract had a potential total value of over $200 billion during the life of the program. The U.S. planned to buy nearly 3,000 JSFs for its air force, navy
-
"twin cities" draw thousands of Mexican people into the already overpopulated region; thus, making the area a major "cesspool". Twin plants were to act as economy boosters for both the US and Mexico; nevertheless, the border
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: "I have a sense of which I can see. I can see that it is a number." and then say it out loud, "112, 123, 134, 145,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
are able to explore moral consciousness and responsibility, as well as themes of charity and giving. Themes of mental health are also explored through extracts such as ‘The Tell Tale Heart’
-
small in the night sky. We sometimes sing the popular nursery rhyme, “Twinkle, twinkle little star….” Actually, however, some of them (like the Sun)
-
osporins. Varying degrees of inhibition of the beta-lactamase with clavulanic acid have been reported. Infections due to multidrug-resistant Gram-negative pathogens in the ICU have invoked the re-emergence of col
-
o. For this reason, technical training and further education is subject to enormous change.⏎Learning trends in basic and advanced training⏎Many studies show that people's attention span has fallen sharply in recent years. This is
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
- L8:60 weakly excites
- L8:1073 weakly excites
- L7:1148 weakly suppresses
- L7:1853 weakly suppresses
- L6:323 weakly excites
- L8:1330 weakly suppresses
- L8:2677 weakly suppresses
- L7:71 weakly suppresses
- L3:1373 weakly excites 'be' after 'to' in infinitives
- L7:851 weakly suppresses
- L7:2530 weakly excites
- L8:816 weakly excites vague "certain" and "such" before a noun
- L4:2817 weakly excites
- L6:1830 weakly suppresses
- L8:815 weakly excites
- L3:606 weakly excites
- L3:2887 weakly suppresses
- L3:1914 weakly suppresses
- L7:2363 weakly suppresses
- L8:1312 weakly suppresses
- L7:2416 weakly suppresses
- L0:2774 weakly suppresses
- L5:2107 weakly suppresses
- L3:2554 weakly suppresses
Feeds later layers
- L14:2188 excites
- L18:1017 suppresses
- L10:1844 weakly excites
- L11:599 weakly excites
- L11:2062 weakly excites
- L10:707 weakly excites
- L15:471 weakly excites
- L15:1870 weakly excites
- L16:1508 weakly suppresses
- L11:2698 weakly suppresses
- L10:2694 weakly suppresses
- L13:523 weakly suppresses
- L13:1400 weakly suppresses "years" after a distance or time number
- L23:995 weakly suppresses
- L16:2714 weakly excites
- L11:38 weakly suppresses
- L24:1072 weakly excites
- L23:2946 weakly suppresses
- L15:1872 weakly suppresses
- L23:2652 weakly excites
- L19:2327 weakly excites
- L17:451 weakly excites
- L16:251 weakly suppresses
- L18:3005 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
- LING
- (mod
- ULO
- simulate
- Forever
- ins
- ulet
- 模拟
- .SelectCommand
- blers
Suppresses
- ��
- внимание
- sehen
- 是否存在
- hài
- :number
- xiety
- nier
- 馁
- awy