the word 'other' after 'each'
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. 3 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 "other" when it comes right after "each", as in "each other" or "from each other". Turned up, it pushed the model toward counting or listing numbers, which is not obviously related.
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.
-
Chinese idioms, 毋忘在莒 strangely has actually two stories of origin, which gave this idiom two different meanings, though somewhat related to each other.⏎The first story of origin came from the early Spring and Autumn period. The
-
, there is a rocky giant called Mt. San Gorgonio. It is 11,502 feet tall—a half a head higher than the other peaks flanking it in the San Bernardino range. By early spring, when most
-
storm? It's good practice for talking, so encourage your baby's efforts by chatting and reading together. Your baby is also figuring out how objects relate to each other and fit together, so it's a perfect time for stacking toys. This week
-
ence electrons from the positive charge of the nucleus.⏎Electron-Electron Repulsions ⏎When two electrons are in the same shell, they will repel each other slightly. This effect is mostly canceled out due to the strong attraction to the nucleus
-
a protein to work ... . For example, if it is the job of one protein to bind specifically to a second protein, then their two shapes must fit each other like a hand in a glove. If there is a positively charged amino acid on
-
<|endoftext|>While humans prize both traits, what’s the real distinction between compassion vs empathy? Though these qualities are often used interchangeably, they’re actually different from each other. To explore the intricacies of these two emotional responses and find out how 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: $12$ $12$ $13$ $14$ $15$ $16$ $17$ $18$ $19$ $20$ $21$ $22$
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
04 pieces.⏎The organisation also aims to allow art enthusiasts to enjoy marvelling at the brilliant and unique Chodiev collection of Kubotas. Jacqueline Atkin, a North Carolina-based art curator and historian, is in charge of the whole
-
first idea that comes to our mind is cutting down on carbohydrate. The world is filled with various fad diets. From keto diet to paleo diet and atkin diet - almost all of these weight loss diets recommend omitting or reducing carbs from
-
, breathing exercises, pranayama (breath control), and self-control. When you have almost any queries concerning wherever along with tips on how to employ
-
34, 1650, 1733, 1703 and 1707 are included under the topic Early Cox History in all our PDF Extended History products.⏎Another 147 words(
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
- L2:36 excites
- L2:49 suppresses
- L2:111 suppresses
- L0:1224 suppresses
- L1:2801 excites
- L2:70 weakly excites
- L0:178 weakly suppresses
- L0:1301 weakly suppresses
- L2:1334 weakly suppresses
- L1:1878 weakly suppresses
- L2:1710 weakly suppresses
- L2:1899 weakly excites
- L2:2328 weakly suppresses
- L2:1494 weakly suppresses
- L2:2771 weakly excites
- L1:630 weakly suppresses
- L0:2821 weakly excites
- L2:75 weakly suppresses
- L2:366 weakly suppresses
- L0:987 weakly excites
- L2:570 weakly suppresses
- L2:237 weakly excites
- L2:126 weakly excites
- L0:1085 weakly excites
Feeds later layers
- L4:1825 suppresses
- L4:849 suppresses
- L4:2515 suppresses
- L6:1047 suppresses
- L6:1525 suppresses
- L9:1464 weakly suppresses
- L4:1302 weakly suppresses
- L7:1133 weakly suppresses
- L9:2174 weakly suppresses
- L6:49 weakly excites
- L9:42 weakly suppresses
- L20:870 weakly suppresses
- L5:3022 weakly suppresses
- L6:1178 weakly excites
- L24:1164 weakly suppresses
- L25:2626 weakly excites
- L4:2406 weakly excites
- L23:167 weakly suppresses
- L15:1056 weakly excites
- L10:2448 weakly excites
- L4:1232 weakly excites
- L8:24 weakly suppresses
- L14:2884 weakly excites
- L22:2164 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
- 物
- ь
- чин
- ее
- ysis
- chron
- minated
- 丐
- суд
- 人造
Suppresses
- auss
- itivity
- rowned
- gende
- grado
- reste
- avenport
- /full
- 背着
- قه