'who' after a person noun
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. 1 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 'who' when it follows a word for a person, like 'a person who', 'someone who' or 'a woman who'. Turning it up did not make the model's writing focus on people, so the output effect is not clear.
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.
-
<|endoftext|>Acute paranoia is a clinically recognized personality disorder characterized by extreme distrust, hostility and suspicion of others for unfounded and often delusional reasons. A person who suffers from paranoia is limited regarding healthy social interactions including
-
.⏎Emotional problems, such as stress, anxiety, or depression, can make a person more forgetful and can be mistaken for dementia. For instance, someone who has recently
-
under copyright without permission, the owner may sue for the value of the violation. Most such cases are handled by civil law. In more serious cases, a person who copies a work that is protected under copyright could be arrested, fined or even go
-
the calorie count of any food;⏎spends more than 2 hours in the gym AND could be underweight? Health-conscious, overzealous, or someone who could have an eating disorder?⏎If you opted for the last one, you are
-
Towns⏎|A religious settlement.⏎|A person who teaches his or her religion to others.⏎|A cowhand; a person who
-
type of shoe that limits pronation or inward rolling of the foot), and running style, which can be influenced by strengthening the hip and core muscles.⏎A runner who increases mileage or intensity too quickly, such as by doing too many speed workouts or
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 thinking about lately, and I have been thinking about lately. I think that is the time I have been thinking about lately. I think that is the time I have been thinking about lately. I think that is the time I
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
itensis and B. abortus disease since 1958 (4).⏎In Bulgaria, until 1998 serologic screening was mandatory for all cattle, sheep, and goats >12 months of age. Afterward,
-
smart devices. From small household appliances to large industrial tools, these objects all have unique identifiers and can communicate with each other. They also have …What is Internet of Things? Read More<|endoftext|>Physicist who slowed the speed of light to speak
-
stories in fact tell us about Franciscan fathers who had been⏎fundamental for all complexes’ building. Some had been the inventors, like Bernardino Caimi
-
. All these caveats make for a murky takeaway.⏎But a lot of other animals eat meat. The article doesn’t talk about our cooking of meat, which I think is the key to how we’ve evolved.<|endoftext|>Is the rising popularity of
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
- L0:1224 suppresses
- L0:178 suppresses
- L0:1301 suppresses
- L0:1389 weakly excites
- L0:1817 weakly excites
- L0:2000 weakly suppresses
- L0:1526 weakly excites
- L0:2345 weakly suppresses
- L0:489 weakly excites
- L0:2250 weakly excites
- L0:1167 weakly excites
- L0:2178 weakly suppresses
- L0:1766 weakly excites
- L0:31 weakly suppresses
- L0:2350 weakly excites
- L0:703 weakly excites
- L0:1118 weakly suppresses
- L0:2549 weakly suppresses
- L0:2481 weakly suppresses
- L0:230 weakly suppresses
- L0:733 weakly excites
- L0:2652 weakly suppresses
- L0:972 weakly excites
- L0:1561 weakly excites
Feeds later layers
- L2:105 suppresses
- L3:402 excites
- L9:865 excites
- L2:302 suppresses
- L7:685 excites
- L6:958 weakly excites
- L14:1798 weakly excites
- L5:2773 weakly excites
- L3:94 weakly suppresses
- L7:2390 weakly excites
- L3:1593 weakly suppresses
- L7:1382 weakly suppresses
- L7:1484 weakly excites
- L10:1124 weakly excites
- L7:173 weakly suppresses
- L21:2185 weakly suppresses
- L2:2116 weakly suppresses
- L9:2442 weakly excites
- L2:560 weakly suppresses
- L14:298 weakly excites
- L2:1984 weakly suppresses
- L15:317 weakly suppresses
- L15:840 weakly excites
- L15:2728 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
- hor
- iors
- ewire
- ensored
- "go
- nels
- ród
- irectional
- 弘
- horizontal
Suppresses
- imum
- vince
- ział
- INESS
- rút
- ใจ
- ւ
- ;)
- ươ
- ibal