indefinite 'a' or 'an' before a 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. 14 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 responds to the small words "a" and "an" right before a noun, especially after verbs or prepositions such as "onto a", "fill a" or "of a". Turning it up produced rambling text that did not clearly echo this pattern.
Written by Claude from this neuron's activations and steering outputs. Confidence: medium. Think it is wrong? Flag it (opens an email to the maker with the neuron already filled in).
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.
-
substance, usually a medication or bioactive agent, transdermally by repulsive electromotive force, through the skin. A small electric current is applied to an iontophoretic chamber placed on the skin, containing a charged active agent
-
Hadid surprised the crowd. The audience shock-factor came from the garment that Bella Hadid wore, which was a spray-on dress. Hadid walked slowly onto a platform in the middle of the runway. Here she was sprayed by two technicians with
-
10,000 people — most of them black — homeless.⏎But in his new book, Pulitzer Prize winning author David Zucchino re-examines a white attack on another black community — in Wilmington, North Carolina — that in many
-
rough shape,” Sam Turner, director of archaeology at the St. Augustine Lighthouse and Museum, said of the button.⏎Even so, the top part of a crown can be seen on the button and similar crowns are found on Royal Provincial
-
power, so it could be used in remote regions where water purity is a concern due to lack of centralized treatment. The user would load the right cartridge, fill a reservoir with the water to be tested, and run the punch card through. This
-
to them.⏎Current Event April 16, 2019⏎The Grand Canyon National Park recently celebrated its 100th birthday. A park ranger there created a “pop-up project,” placing an old typewriter on an
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 need to be able to think about the role of a system, and then I need to be able to think about the roles, and then I need to be able to respond to the roles, and then the roles should be connected to
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
economy.⏎In order to find a solution to the current financial situation, it is necessary to understand the actual problem. It is insane for one to expect to solve a problem with incorrect information. Therefore we need to end some myths regarding the budget.⏎
-
departments, universities, etc.) having jurisdiction over disposal of sewage, industrial wastes, stormwater or other wastes. This includes special districts under state law, such as a sewer district; flood control district, drainage districts or similar entity; an Indian tribe
-
. (Celsius)||-65|⏎|Max. Temp. (Celsius)||165|⏎There are no reviews yet.<|endoftext|>Talk about doing something in the blink of an eye.⏎Google wants to take that phrase to a new
-
to take risks, and was hard headed once he had made up his mind on something. At the time it was unclear how much oil was left in the ground and
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
- L11:331 strongly suppresses
- L11:875 strongly excites
- L10:1377 strongly excites
- L4:49 strongly excites
- L10:120 strongly excites
- L13:1234 strongly excites
- L9:662 strongly excites
- L10:1162 strongly suppresses
- L4:73 excites
- L13:824 excites
- L11:1168 suppresses
- L2:205 suppresses
- L6:426 excites
- L4:740 excites
- L5:225 suppresses
- L5:705 suppresses
- L5:153 excites
- L9:2198 suppresses
- L12:1868 suppresses
- L11:2859 suppresses
- L11:1245 excites
- L7:155 suppresses
- L9:24 suppresses "the" before a named organisation or title
- L12:966 suppresses
Feeds later layers
- L18:2630 excites
- L15:1888 excites
- L17:2190 suppresses 'the' starting an object phrase
- L18:1547 suppresses
- L15:3031 excites
- L16:2248 excites
- L17:1232 suppresses
- L17:2288 excites
- L17:1816 suppresses
- L17:2011 suppresses
- L19:1543 suppresses
- L17:1652 excites
- L17:2202 excites
- L16:2861 suppresses
- L16:2062 suppresses "other" and contrast words
- L15:2402 suppresses
- L20:1983 suppresses
- L15:2571 weakly suppresses
- L26:2817 weakly excites
- L15:1849 weakly suppresses
- L16:2466 weakly suppresses
- L17:188 weakly excites 'a' before a noun phrase
- L16:1212 weakly suppresses
- L18:1449 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
- 有
- odo
- thes
- 有不同的
- 常见的
- Open
- 好评
- 新闻记者
- 国立
- ODO
Suppresses
- ��
- ROLS
- ::$
- ";}⏎
- $"
- RuleContext
- KANJI
- 陛下
- stuff
- _parms