unclear: list items and everyday prepositions
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. 14 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. 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 words that finish an everyday phrase, like 'stay close', 'rest your head on', or the 'nuts' in a list of foods. The link is loose, and turning it up didn't produce a clear theme.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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.
-
clothes with someone you trust.⏎DURING A VIOLENT INCIDENT⏎• Call 911 or local police.⏎• If an argument starts, stay close to an exit.⏎• Stay away from the bathroom, kitchen, or any other
-
/1<|endoftext|>Some may argue that sleeping is one of the best parts of the day. You get into bed, snuggle under the covers, rest your head on a comfortable cushion, and switch off your thoughts for the
-
<|endoftext|>Remember snuggling with your child, book between you, the scent of baby shampoo under your nose? Most of us think such moments are gone after our snuggler learns to read. But they don’t have to be! Save the ritual
-
solve their problems.⏎Read also, 10 lines short stories for kids.⏎2. Hungry dogs⏎Three dogs lived in a city. They were very close friends. One day, all three dogs were very hungry.⏎They started searching for
-
are potato chips, French fries, pop, red meat and processed meat.⏎Foods that do not contribute to weight gain are fruits and vegetables, whole grains, nuts and yogurt.⏎This is based on following 120,000
-
for the day. If you have time constraints, consider options such as protein smoothies, healthy muffins, hardboiled eggs or granola with fruit, nuts and seeds.⏎Eliminate or Min<|endoftext|>Venezuela’s human rights record was
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 been a a contradiction in my own life. I have been a person who did not exist, but I am now... I have been a person, I have been a person, I have been a person. I have been a
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
1883 we were walking along past a large field of⏎potatoes in North Oxford, where we beheld the strange spectacle of a⏎pair of bipeds drilling their offspring to march up one potato row and⏎down the other
-
have to illustrate the sentence on a large piece of paper in front of the class, while the rest of the class guesses what sentence it is. The first one to correctly say the sentence is the next one to choose a sentence and do the illustration
-
fire and air.⏎He was a strong proponent of the sustainable use and cultivation of land and water, proper treatment of animals, plants and birds, and the equal rights of users. In this context the modernity of the Prophet sall Alla
-
taught personal responsibility through leading by both word and deed. Yet lacking an in-home role model or role models -- my father grew up as an out-of-wedlock only child -- does not necessarily determine one's fate or destiny. Sometimes one must
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
- L10:1270 weakly excites
- L12:2148 weakly excites
- L3:2994 weakly suppresses
- L10:1968 weakly excites
- L7:2317 weakly suppresses
- L10:2243 weakly excites
- L13:397 weakly excites
- L12:1615 weakly suppresses
- L12:1435 weakly excites
- L7:2785 weakly excites
- L8:2299 weakly suppresses
- L10:2038 weakly excites
- L7:2814 weakly excites
- L9:760 weakly suppresses
- L2:1264 weakly suppresses
- L12:2779 weakly suppresses
- L10:2637 weakly suppresses evaluative words: good, bad, positive
- L8:997 weakly excites
- L8:1497 weakly suppresses
- L5:724 weakly suppresses
- L6:1961 weakly suppresses
- L8:1449 weakly suppresses
- L8:899 weakly suppresses
- L8:229 weakly excites
Feeds later layers
- L22:983 weakly excites
- L19:2312 weakly suppresses
- L25:2313 weakly suppresses
- L18:1175 weakly excites
- L24:1394 weakly excites
- L22:1947 weakly suppresses
- L20:2454 weakly excites
- L20:1653 weakly suppresses
- L20:1625 weakly excites
- L27:1961 weakly excites
- L15:2809 weakly excites
- L21:1325 weakly suppresses
- L23:2084 weakly excites
- L20:2626 weakly suppresses angle and right-angle phrases
- L19:2817 weakly excites
- L24:1713 weakly excites
- L20:1035 weakly excites
- L23:1367 weakly suppresses
- L17:1074 weakly excites
- L21:234 weakly suppresses
- L21:1159 weakly suppresses
- L20:1509 weakly suppresses
- L19:1097 weakly suppresses
- L22:1325 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
- 为期
- igte
- 捣
- -response
- ัส
- Inspectable
- _food
- weight
- -track
- 饼干
Suppresses
- 开放
- vampires
- 败
- 敗
- .disconnect
- .twitch
- inema
- cuts
- célib
- yle