'those' standing in for a noun in comparisons
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 lights up on the word "those" (or "one") when it stands in for a noun in a comparison, as in "better than those of other countries." When it was turned up, the model's writing drifted into rambling, unrelated text, so the effect on its output was not clearly readable.
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.
-
More Info... by email⏎Consumer ethnocentrism is a psychological concept that refers to individuals who believe that their country's products are superior to those of other countries. This concept also describes consumers in one country thinking that purchasing products
-
are more "right-beaked" crossbills than left-beaked? Birds that have the lower mandible crossing to the right are three times more common than those with a left-crossing lower mandible.⏎- White-winged Crossbills
-
magnetic pole. They find that over a short period, possibly only a few centuries, the pole leapt from a location oriented around 45 degrees south to one near 45 degrees north. The paleomagnetic pole then drifted through around
-
, right? See it in action below, and check out the project’s elaborate overview here.<|endoftext|>Our system of government is different from those of all other nations, because part of the political power is vested in the State
-
an odd shape in the mutant mice: Their axons and dendrites are shorter and have fewer branches than those of controls. The shortage of branches may translate to fewer synapses, or neuronal
-
conscientiousness and signs of the disease. The brains of the conscientious had similar levels of plaques and tangles which build up and cause Alzheimer's as those of the less conscientious.⏎Dr Wilson said this was not unexpected. He said
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 current and the current's current, and the current's current's current's current's current's current's current's current's current's current's current's current's current's current's current's current's current's current's current's
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
, began to terrorize Iraqi cities and towns. Al-Qaida in Iraq recruited from other countries to build its influence and keep a democratic nation from arising in the region.⏎As the challenges of helping to form a free Iraq began to
-
soldiers and enemies, is fraught with danger. Along the treacherous path to Belgium the three children skate using every bit of speed, courage, and strength they can muster. All the time they
-
be one of the largest in the area. For surrounding landowners, the project will promote different methods of long-term forest
-
around 400 CE. The oldest inscription is written in faulty Sanskrit. After this, inscriptions alternate between Sanskrit and the Cham language of the times.⏎Cham kings studied classical Indian texts such as the Dharmaśāstra
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
- L13:2436 excites
- L6:345 weakly excites
- L7:1179 weakly suppresses
- L12:575 weakly suppresses
- L8:1001 weakly suppresses
- L13:2983 weakly suppresses
- L13:1173 weakly excites
- L8:1990 weakly suppresses
- L6:2276 weakly excites
- L13:533 weakly suppresses
- L10:1329 weakly suppresses
- L12:1290 weakly excites mid-word fragments in names
- L12:1047 weakly suppresses
- L10:583 weakly excites
- L6:2417 weakly suppresses
- L1:1794 weakly excites
- L1:2119 weakly excites
- L2:1781 weakly suppresses
- L4:390 weakly excites "and" joining abstract nouns
- L13:2796 weakly excites
- L13:44 weakly excites inside web addresses and citations
- L1:2539 weakly excites
- L6:924 weakly excites
- L11:546 weakly suppresses
Feeds later layers
- L15:851 excites
- L16:662 suppresses
- L15:1902 excites
- L17:177 excites
- L18:68 excites
- L15:795 excites
- L17:76 excites
- L17:1918 excites
- L20:284 excites
- L16:2715 excites
- L17:1167 excites
- L18:1143 excites
- L18:579 excites
- L15:733 suppresses
- L19:211 excites
- L17:181 excites
- L21:665 excites
- L20:55 suppresses
- L17:2037 excites
- L15:1601 suppresses
- L15:894 weakly excites
- L15:1589 weakly excites
- L16:656 weakly suppresses
- L15:232 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
- onis
- ux
- of
- igation
- favor
- iges
- aus
- den
- tô
- ationToken
Suppresses
- ASIC
- /DTD
- serialVersionUID
- cala
- imensional
- 或多
- 小时内
- .UnitTesting
- (QObject
- 的研究