'because' introducing a reason
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. 13 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 words like 'because' or 'is' right after a claim that is about to be explained, such as 'the reason for this is'. When turned up, the model drifted into odd, unrelated text, so the steering does not clearly match.
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.
-
are several reasons why people may choose to eat white turkey meat. Some people believe that it is more tender and flavorful than dark meat. Others prefer the white meat because it is lower in fat and calories. And some people simply like the way it
-
out of principle alone. For starters, he refused to ever short a stock. That’s because shorting a stock is basically betting on someone else to fail, meaning you profit
-
way the decision goes.⏎Those are the ‘easy’ situations<|endoftext|>You hear it all the time, “don’t write your picture book in rhyme.” That’s because many writers don’t do it well. The story suffers to fit the rhyme;
-
their nests and babies.⏎● To help them fly. Geese honk while they’re flying to help them maintain their V formation. They fly in this formation because it helps them cut down on wind resistance, allowing them to save energy.⏎●
-
In addition, 50 percent more men are affected than women, according to the National Institute of Neurological Disorders and Stroke. However, the reason for this is unclear.⏎Family history is another important risk factor. Individuals with a parent or sibling
-
relor.⏎Counseling Point for Anti-Platelets:⏎Aspirin and warfarin are sometimes prescribed together, and this is because one is anti platelet and the other is anticoagulant. Please see
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: 10000000000000000000000000000000000000000000000
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
ex-ante consumption expenditure is Rs. 250 Cr. and the ex-ante investment expenditure is Rs. 100 Cr. in an economy without a government.⏎21. Differentiate between foreign trade surplus and
-
osacral portion) or lower extremities.⏎Causes of chronic pelvic pain⏎Chronic pelvic pain may originate due to multiple etiological factors. The commonly studied causes of chronic pelvic pain are as follows-⏎- Central and peripheral nervous systems:
-
the occasion will be celebrated as a sign of the boy's passage into maturity⏎and promising future. A significant event in the culture and history of South⏎India is known as the Dhoti Ceremony. This is a momentous day in the
-
joint project of the non-profit National Association of County and City Health Officials (NACCHO) and the Centers for Disease Control and Prevention (CDC), the U.S. government agency that “serves as the national focus for developing and applying disease
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
- L12:2880 suppresses words that expect an of, to or for next
- L12:124 suppresses
- L12:2340 excites
- L11:1793 excites
- L10:614 suppresses
- L11:2997 suppresses
- L12:1760 suppresses
- L7:439 excites
- L12:2402 excites
- L11:305 excites
- L12:811 excites
- L7:2811 excites
- L9:2052 excites 'because' and 'that' after a reason
- L12:2779 weakly excites
- L11:1465 weakly suppresses
- L6:431 weakly excites
- L11:1986 weakly suppresses
- L10:1977 weakly excites
- L11:383 weakly suppresses
- L12:292 weakly suppresses
- L7:1307 weakly suppresses
- L11:2087 weakly excites
- L12:347 weakly suppresses
- L10:331 weakly excites
Feeds later layers
- L14:1096 suppresses
- L15:34 excites
- L14:1503 excites
- L17:24 excites
- L14:377 excites
- L17:2264 weakly suppresses
- L16:1474 weakly suppresses
- L22:612 weakly suppresses
- L22:1730 weakly suppresses
- L19:545 weakly excites
- L15:2765 weakly excites
- L14:2032 weakly excites
- L17:905 weakly suppresses
- L14:555 weakly excites
- L16:2677 weakly excites
- L14:735 weakly suppresses
- L16:1067 weakly excites
- L15:2589 weakly suppresses
- L19:865 weakly excites
- L16:1599 weakly suppresses
- L14:3009 weakly suppresses
- L14:1741 weakly suppresses
- L17:110 weakly excites
- L14:2003 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
- ери
- VERT
- Jenner
- STER
- Vance
- 元左右
- Piper
- mer
- 往
- ницы
Suppresses
- %=
- MonoBehaviour
- 髑
- ksi
- bases
- .Generated
- jede
- extends
- 蹇
- spins