'way' in 'work their way' phrases
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. 4 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 lights up on the word 'way' in phrases like 'work their way up' or 'find its way'. Its negative side points to traffic lights and roundabouts. Turning it up did not produce writing that clearly connects to these phrases.
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.
-
throughout the world and will often latch on to shoes or clothes, and then work their way up your clothing until they find a prime piece of exposed skin. When they find
-
in more efficiently, and professors are able to use more of their time with students individually as their office schedules are easier to manage. Many students have to work their way through college to support themselves while receiving an education. This is much easier with a
-
others for survival. They would rather work for meagre salary for now and work their way to higher salaries or income positions. Those who believe they are owed something may never
-
policy.⏎The Magic School Bus Lost in the Solar System Unit Study⏎My kids are IN LOVE with the Solar System right now. So we have been working our way thru our The Magic School Bus Lost in the Solar System Unit Study. We have
-
the image of a Native American find its way to Ghent? And how should we deal with it when placed within a modern
-
ethical in a given situation if the problem is simple and straightforward. But when a complex dilemma leaves you uncertain, there are some tools that can help you work your way to a good decision. They can’t tell you what to do, but they
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 green light, and there is no way. The green light is green. The green light is red. The green light is red. The green light is red. The green light is red. The green light is red. The green
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
The standard Teletype paper tape uses only 8 bits.) Cities would hire private companies to translate the data into reports that traffic engineers could use to adjust traffic lights or improve roads.⏎Bill Gates and Paul Allen were high school students at Lakeside
-
around a detailed communication protocol, which we also present. We demonstrate in simulation that our new mechanism has the potential to significantly outperform current intersection control technology -- traffic lights and stop signs. Because our mechanism can emulate a traffic light or stop sign,
-
deciding on whether to turn left or right at an intersection, whether to slow down when approaching a roundabout, or to stop for a pedestrian at a pedestrian crossing. A fully automated vehicle
-
ten objects from museum collections in Greater Manchester⏎A compass used in a PoW escape, a railway station sundial and part of the world's first commercial computer: objects that tell a
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
- L3:1216 suppresses
- L3:294 weakly excites
- L1:3020 weakly suppresses
- L0:2338 weakly suppresses
- L3:115 weakly suppresses
- L3:2133 weakly excites
- L3:2906 weakly excites
- L3:1514 weakly suppresses
- L3:322 weakly suppresses
- L3:2437 weakly suppresses
- L2:1161 weakly suppresses
- L3:187 weakly excites
- L1:753 weakly excites
- L1:2650 weakly suppresses
- L3:703 weakly excites
- L1:1826 weakly suppresses
- L0:1301 weakly suppresses
- L3:567 weakly excites
- L2:549 weakly suppresses
- L0:1224 weakly suppresses
- L0:178 weakly suppresses
- L3:1327 weakly suppresses
- L3:2140 weakly suppresses
- L3:691 weakly excites
Feeds later layers
- L6:2200 suppresses
- L6:2898 suppresses
- L7:1674 weakly suppresses
- L10:2160 weakly suppresses
- L22:1675 weakly excites
- L5:1805 weakly suppresses
- L5:2591 weakly excites
- L6:2762 weakly suppresses
- L5:2070 weakly suppresses
- L5:2145 weakly suppresses
- L6:1226 weakly excites
- L7:95 weakly suppresses
- L6:1881 weakly suppresses
- L6:1471 weakly suppresses
- L16:2029 weakly suppresses
- L18:1592 weakly suppresses
- L13:2536 weakly excites
- L10:2919 weakly excites
- L21:2718 weakly suppresses
- L11:2962 weakly excites
- L5:1202 weakly suppresses
- L5:2617 weakly suppresses
- L15:393 weakly suppresses
- L7:2255 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
- 茵
- itemName
- 孢
- anded
- 钽
- 浸泡
- WithEmail
- 与中国
- .xmlbeans
- entrant
Suppresses
- 心动
- 闸
- ngừng
- kenn
- 買う
- 見る
- zen
- 례
- bie
- compr