capital V starting a word
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. 27 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 a capital 'V' that starts a word in the middle of a sentence, like 'Vaping', 'Vihara', or 'Vascular'. Its opposite is a lowercase 'v' starting a word, like 'vases'. Turning it up or down gave text that did not clearly relate to this pattern.
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.
-
also impair memory and concentration. Nicotine can also affect mood and impulse control.⏎Although vaping may be safer than using cigarettes, there are still potential health risks. Vaping devices are still new and researchers have not been able long-term to examine the
-
⏎In 1993, the foundations to the Dhamma Talaka Peace Pagoda were laid in Birmingham, England. Supported by the Birmingham Buddhist Vihara, the pagoda underwent a large amount of planning. The circumference of
-
I. Book I. Blood: The Tissues of Movement, the Vascular Mechanism. Sixth Edition. 10s. 6d.
-
Achilles Tendon Tear.⏎Rotator<|endoftext|>When you work with RAID 5 volumes, consider the Requirements for RAID 5 Volumes and Guidelines for RAID 5 Volumes. Many striping guidelines also apply to RAID 5 volume configurations. See
-
ated with fat, fibrous and elastic tissue⏎- Increases circulation by pushing blood out of foot⏎What are arteriovenous anastomoses?⏎- Vascular connections in the corium⏎- There are smooth muscles at arterial end;
-
are because of early menopause. However, menstruation cycle changes can also occur because of sudden change in lifestyle and lack of nutrition.⏎Less Lubrication Of Vagina⏎A drop in the estrogen levels in the body during menopause may lead
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: why did people person person person person person person person person person person person person person person person person person person person person person person person person person person person person person person person person person person person person person person person person person person person person person
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
the life of cut flowers.⏎- Always keep vases and flower containers clean, clean, clean. Bacteria-laden water or
-
of the arteries with plaque build up (artherosclerosis), very fast heart rate, very high blood pressure or low heart rate or low blood pressure, drugs causing vaso constriction such as cocaine, smoking, methamphetamine, and
-
create problems. Field guides — even excellent new ones like Sibley or National Geographic — show 'standard' individuals. There is excellent literature about the Warbling v. Philadelphia Vireo problem, and more observers should review this literature before claiming
-
ing parent. As a result, they might have learned to suppress their anger for fear of retaliation or made a personal vow never to be like their parent.⏎The venter style is at the opposite end of the anger management continuum. Unlike the suppress
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
- L22:300 strongly suppresses
- L20:336 strongly suppresses
- L18:423 suppresses
- L23:2815 suppresses
- L18:365 excites
- L17:765 excites
- L2:2142 suppresses
- L0:2419 excites
- L5:1712 suppresses
- L1:18 suppresses
- L21:2454 weakly suppresses
- L6:13 weakly excites
- L10:757 weakly suppresses
- L18:778 weakly suppresses
- L17:389 weakly excites
- L16:95 weakly suppresses
- L5:3052 weakly suppresses weighing, counting and measuring
- L8:4 weakly excites
- L6:12 weakly excites
- L19:974 weakly suppresses
- L7:421 weakly excites
- L1:972 weakly suppresses
- L5:2394 weakly excites
- L26:1176 weakly excites
Feeds later layers
None: this is the last layer.
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
- 讓他們
- setState
- ところ
- setState
- Mixer
- ary
- State
- 几家
- .Convert
- uD
Suppresses
- ega
- .self
- ik
- 基数
- doubles
- Ử
- 配偶
- ander
- vastly
- uples