'to' after 'too big' or 'right angles'
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 responds to the word 'to' when it follows an adjective or a phrase like 'too big', 'too light', or 'at right angles', as in 'too big to be human'. It is most strongly pushed down by 'to' in phrases about getting somewhere, like 'all the way to the top'. Steering did not give a clear theme.
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.
-
defined in terms of its⏎(i) Isometric System :⏎Three axes of crystal are⏎of equal length and lie at right angle to each⏎other, e.g. Halite mineral.⏎(ii) Tetragonal
-
were satisfied with these explanations. Some radar operators and pilots insisted that the movements of the objects were too coordinated to be random atmospheric anomalies. Furthermore, skeptics pointed out that temperature inversions were
-
pure cinnabar sources. When sold in Europe, it was frequently adulterated with red lead, an inexpensive bright lead oxide pigment that was too reactive to be trustworthy enough for use in art. Vermilion was favored by old paint masters
-
see this for yourself, stick two fridge magnets together, back to back. They will slide against each other in one direction. If you slide them at right angles to that, they will alternately attract and repel one another.<|endoftext|>E.
-
.⏎A couple of days later, police and investigation officers showed up on their door, as per a local news report. They determined that the bones were too big to be human, and told the couple that it was of
-
axes is perpendicular⏎to other axes and is of different length, e.g.⏎(iv) Orthorhombic System :⏎Three axes intersect⏎at right angles to each other, but they⏎are of unequal lengths, e.g. Stau
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 can see a lot of the point of the goal, and I am able to reach it. I can feel it. I have reached the goal. I can feel it. I can feel it. I have reached the goal. I
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
a meteor passes through the Earth's atmosphere at high speeds, it heats up, glows brightly and finally vaporizes. If a meteor makes it all the way to the ground, it is called a meteorite.⏎Meteors occur randomly during
-
. Even if you don't make it all the way to the top of the mountain or hike all 40 miles of the trail,
-
is up to, and she slams the window sash on his beard. Trapped, he must surrender the ladder.⏎After climbing to the top of the mountain, she releases him by pulling on a long rope.
-
during the germination period (usually 1 - 2⏎weeks) by applying frequent, but light, irrigations.⏎Large lawns or hard-to-reach areas may be better hydroseeded.⏎Pattern for sowing seed<|endoftext|>MONT
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
- L3:1859 suppresses
- L2:872 excites
- L2:750 weakly excites
- L3:592 weakly suppresses
- L3:59 weakly excites
- L3:50 weakly excites
- L0:1456 weakly suppresses
- L3:2267 weakly suppresses
- L3:2179 weakly suppresses
- L2:2875 weakly excites 'up' after show, turn and showed
- L3:1181 weakly suppresses
- L3:1383 weakly suppresses
- L2:1919 weakly excites
- L2:1321 weakly suppresses
- L2:1174 weakly suppresses
- L3:272 weakly excites
- L3:122 weakly excites
- L2:2956 weakly suppresses
- L3:115 weakly suppresses
- L2:2304 weakly suppresses
- L1:906 weakly suppresses
- L3:127 weakly excites
- L2:1148 weakly suppresses
- L2:1036 weakly excites
Feeds later layers
- L26:1264 suppresses
- L8:2289 suppresses
- L8:1627 weakly suppresses
- L16:2690 weakly suppresses
- L7:1883 weakly suppresses
- L23:1277 weakly suppresses
- L25:1528 weakly suppresses
- L8:1716 weakly suppresses
- L17:2282 weakly suppresses
- L9:723 weakly suppresses
- L16:2744 weakly suppresses
- L16:2027 weakly suppresses
- L26:676 weakly excites
- L26:3055 weakly suppresses
- L5:1903 weakly excites
- L16:1168 weakly suppresses 'to' after verbs of change
- L13:2597 weakly suppresses
- L23:2455 weakly suppresses
- L11:2700 weakly excites
- L17:1369 weakly suppresses "of" after is/are/be (of + quality)
- L6:1149 weakly suppresses
- L10:2980 weakly excites
- L7:2555 weakly excites
- L14:1599 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
- visto
- izes
- dụ
- Knife
- Wrapper
- Int
- fires
- Wrapper
- гон
- قر
Suppresses
- 设计理念
- reached
- reach
- 抵达
- 到达
- reachable
- reach
- reaching
- .sf
- 终点