"the" or "of" before a noun phrase
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. 8 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 small connecting words like "the" and "of" that come right before the thing being talked about, for example "bracing of the leg" or "the continuous effort of the foundation". On the negative side it fires on "of" after a noun, as in "floor of the mouth". Steering did not give a clear theme, so the pattern is only a rough guess.
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.
-
do this:⏎- Start by analyzing the message you want to convey to the reader. Before presenting it, define the problem described in the text, the expected reaction of the reader to the text, the purpose of the text, as well as the
-
software, novel materials, more dexterous robots, new processes (notably three-dimensional printing) and a whole range of web-based services. The factory of the past was based on cranking out zillions of identical products: Ford famously said
-
he spotted a tub of yeast in the room. Recalling an old tradition that rotting meat suspended over yeast would become pure and sweet again, he dosed the youngster with it – and cured him.⏎He treated several other parishioners in
-
test method of the improved Godron M. The plant community stability was closest to the point of stability
-
0.93.3⏎To perform the test, the patient is instructed to stand on one leg without support of the upper extremities or bracing of the unweighted leg against the stance leg. The patient begins the test with the eyes
-
double strand break DNA break (DSB) most often results in the activation, by phosphorylation (indicated in this figure by a circled p), of the PI3 kinase ATM. Multiple pathways are downstream of ATM including
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 number of the days you are in the room, the number of the days you are in the room, the number of the days you are in the room, the number of the days you are in the room... the number of the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Despite being the largest pair of glands, only approximately 25% of saliva is produced by the glands.⏎The are a pair of glands located beneath the floor of the mouth, superior to the . The secretion produced is a mixture of both and
-
triangle is equal to the sum of the squares of the sides containing the right angle." But Pythagoras had an all-inclusive
-
gen cyanide is thought to be the main anticancer compound formed from laetrile when broken down in the body. Amygdalin is found in the pits of many fruits and in raw nuts. It is also found in other plants such as
-
of the Prophet Jonah, “As Jonah was in the belly of the whale three days and three nights, so also shall the Son of Man be in the heart of the earth three days and three nights” (Mt. 12:
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
- L4:1686 strongly excites
- L6:2664 strongly excites
- L5:2454 excites
- L6:357 suppresses
- L6:388 excites
- L6:487 suppresses
- L6:1735 excites
- L5:363 excites
- L6:1114 excites
- L1:1565 excites
- L6:2326 suppresses
- L4:4 suppresses
- L7:8 excites
- L7:23 suppresses
- L6:925 suppresses
- L5:453 suppresses
- L1:752 excites
- L6:1637 excites
- L6:1534 suppresses
- L3:315 excites 'of' after a noun phrase
- L6:2785 excites
- L5:38 suppresses
- L5:1389 weakly excites
- L7:751 weakly suppresses
Feeds later layers
- L10:272 excites
- L10:362 suppresses
- L18:520 suppresses
- L14:1519 suppresses
- L10:469 suppresses 'of' after a noun phrase
- L14:1659 suppresses
- L18:183 excites
- L16:2612 suppresses
- L15:632 suppresses
- L13:203 suppresses
- L15:349 suppresses
- L9:283 excites
- L9:335 suppresses 'of' after nouns like anniversary or onset
- L14:655 suppresses
- L17:865 suppresses
- L9:592 suppresses
- L18:381 suppresses
- L15:1028 suppresses
- L16:1551 suppresses
- L20:482 suppresses
- L12:2849 suppresses '-based' after a place name
- L12:850 excites
- L14:126 excites
- L13:210 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
- ][-
- 在香港
- 是我的
- 让消费者
- 凭什么
- depuis
- 让她
- 是韩国
- 帮我
- нстру
Suppresses
- mankind
- implements
- ELSE
- lic
- adar
- (savedInstanceState
- vik
- ETO
- usra
- unicorn