the word 'average' before a quantity
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. 4 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 'average', especially in phrases like 'an average of 10 percent' or 'an average teacher'. When it was turned up, the model kept writing about averages and computing them.
Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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.
-
group.⏎The results of the studied showed that the groups with the most weight loss were the diet-only and diet and exercise group with women in those groups losing an average of 10 percent of their starting weight, according to the study.⏎Additionally
-
year.⏎Salary by Employment Sector⏎As of 2011, the highest-paid electricians were those who taught at business schools. These workers earned an average of $79,250 a year. Most electricians worked as
-
data were derived from national food consumption surveys, including more than 1000 individuals by country (women and men/18-64 years). Average consumption, GHGE (gCO2eq) and nutritional composition of 1
-
and weeding torches calories, too. A 154-pound person burns an average of 330 calories per hour through gardening, according to the US Centers
-
point) they hosted. Overall, 57.9% of trees in the study sites carried at least one liana. The infested trees hosted an average of 1.
-
of effectiveness) each year produces over $350,000 more income for his/her students compared to an average teacher. On the the other hand, a poor teacher (one in the bottom
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 problem is the same as the set of possible ones? So, what's the problem with the set of possible ones? Let's say that the problem is the same as the set of possible ones? So, what's the problem?
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
:⏎Top Suggestions Vocabulary Worksheet Mix :⏎Vocabulary Worksheet Mix Whether you re planning a summer lesson or
-
tools and materials of high quality came into use, masterpieces of calligraphic art were produced, e.g., the Irish Book of Kells (8th cent.; Trinity College, Dublin; see under Ceanannus Mór) and
-
has been consistently practiced as a major aesthetic expression. In China, from the 5th cent. B.C., when it was first used, calligraphy has always been considered
-
22, at 888-835-5322 (TTY) or at email@example.com.<|endoftext|>Table of Contents :⏎Top Suggestions Senior Kindergarten Writting Worksheet :⏎Senior Kindergarten Writting Worksheet In
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
- L3:2163 suppresses
- L2:2862 excites
- L3:2897 excites
- L3:2291 suppresses
- L3:2686 weakly excites
- L2:356 weakly excites
- L2:2692 weakly excites
- L1:1464 weakly excites
- L3:50 weakly excites
- L1:2801 weakly excites
- L0:2849 weakly excites
- L2:84 weakly excites
- L3:421 weakly suppresses
- L3:3054 weakly suppresses
- L1:930 weakly excites
- L3:1755 weakly suppresses
- L3:2875 weakly suppresses
- L3:115 weakly suppresses
- L3:272 weakly excites
- L3:2047 weakly excites
- L3:59 weakly excites
- L2:1657 weakly excites
- L0:1301 weakly suppresses
- L1:2683 weakly excites
Feeds later layers
- L16:2836 excites
- L6:957 excites
- L8:1463 excites
- L21:1242 excites
- L7:2826 excites
- L16:1623 excites
- L8:2723 weakly excites
- L8:2150 weakly excites
- L11:1534 weakly excites
- L7:795 weakly suppresses
- L6:1153 weakly excites
- L5:1084 weakly suppresses
- L5:1656 weakly excites 'of' after end, top or bottom
- L8:910 weakly suppresses
- L16:1810 weakly excites
- L11:2115 weakly suppresses
- L22:2977 weakly excites
- L11:1160 weakly excites
- L6:674 weakly suppresses comma after a 'said' or 'answered' verb
- L10:503 weakly excites
- L5:811 weakly suppresses
- L8:2846 weakly suppresses
- L6:2961 weakly suppresses
- L17:1445 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
- emme
- ]->
- -sama
- point
- Москве
- 散布
- embro
- ]][
- ensity
- ocus
Suppresses
- zá
- ivers
- 不完
- Patriot
- KR
- ران
- ponible
- 国王
- idon
- ват