'with' after 'provide us' or 'provided you'
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. 9 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
The neuron fires on the word 'with' when it follows phrases like 'provides us with' or 'provided you with'. Its negative side is weak and hard to read. Turning it up gave wandering text that did not clearly relate to providing things.
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.
-
, hypermilers seek to push their gas tanks to the limit and achieve hypermileage, exceeding EPA ratings for miles per gallon.⏎So, there you have it. How long 'till hyper is replaced by wondermileage or excessiv
-
, and provides us with a powerful foundation to ask fundamental questions about the behavior of biological materials.⏎Moreover,
-
the chaos destroying the world is to solve a 2,000 year-old mystery: Who is the Cartographer?More⏎We've sent an email with your order details. Order ID #:⏎
-
the early Mercury missions beginning in 1961. Today, they observe the Earth from their unique point of view aboard the International Space Station, providing researchers with key data and recording changes over time from human-caused changes like urban growth and
-
secrets in the coming months. “This find has the potential to be one of the most significant and exciting archaeological discoveries in the city for many years, providing us with yet more clues as to what life was like in medieval Edinburgh,” he said.⏎
-
, they calculate the surface area of the prism. This provides students with the opportunity to work with complex shapes to find surface area in a given context.⏎
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 end. I think about people and places. I have a feeling that it's been (2) minutes. The distance between A and B is 10 cm. When the train passes the station, 30 cm apart,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
). He erected an altar to Zeus in the Temple in Jerusalem, and sacrificed pigs there, which are unclean to Jews. The Maccabee family led
-
asted brass buttons in the front, and three brass buttons on each cuff. The shoulders are intricately decorated with epaulets bearing silver oak leaves on a blue field, indicating the Infantry branch of the United States' Army. This type of
-
. God gave Moses the Law while Israel wandered in the desert. It was rigid. There was little flexibility in its demands. It declared “an eye for an eye and a tooth for a tooth.”⏎God promised Abraham a son who would be the
-
aru Takei, of Osaka Prefecture University in Japan.⏎Whiskers are considered tactile sensors for specific animals like cats, according to the study. Whiskers found on a cat's face for example can help the animal navigate around objects.⏎
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
- L2:2568 weakly excites
- L6:2457 weakly suppresses
- L7:2208 weakly excites
- L7:1950 weakly excites
- L8:571 weakly excites
- L7:1923 weakly excites
- L4:411 weakly excites
- L3:2533 weakly suppresses
- L6:980 weakly suppresses
- L8:2463 weakly suppresses
- L3:2368 weakly excites
- L1:3036 weakly suppresses
- L3:699 weakly excites
- L4:2558 weakly excites
- L8:233 weakly suppresses
- L4:334 weakly suppresses
- L6:850 weakly excites
- L8:2575 weakly excites
- L8:2531 weakly excites
- L5:459 weakly excites
- L6:1303 weakly suppresses
- L3:1144 weakly suppresses
- L3:1002 weakly suppresses
- L7:2712 weakly suppresses
Feeds later layers
- L10:886 weakly excites
- L24:258 weakly excites
- L23:3022 weakly suppresses
- L11:2015 weakly excites
- L15:1595 weakly suppresses
- L19:2956 weakly suppresses
- L10:1126 weakly suppresses
- L18:604 weakly excites
- L18:2419 weakly excites
- L18:2117 weakly suppresses
- L17:2863 weakly excites
- L10:2288 weakly suppresses
- L25:1147 weakly suppresses
- L17:2009 weakly excites
- L22:2537 weakly excites
- L20:1908 weakly suppresses
- L15:919 weakly excites
- L21:1441 weakly excites
- L12:273 weakly excites
- L12:550 weakly excites
- L22:1496 weakly suppresses
- L10:1911 weakly suppresses
- L17:2956 weakly suppresses
- L21:1051 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
- Churchill
- 给
- 不足以
- 可供
- (provider
- providing
- Patterns
- ning
- illumin
- 多种
Suppresses
- schlä
- amaño
- merg
- acked
- ić
- ��
- ibaba
- maken
- anceled
- ersed