the word 'age' in age groups
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. 23 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 the word 'age' when it means a group of people, as in 'this age group' or 'older age'. Turning it up made the model repeat the word 'group' over and over, which matches what the neuron pushes toward.
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.
-
2009, 18- to 54-year-olds accounted for 95 percent of all Gulf War-era II veterans, whereas that same age group made up about 7 in 10 nonveterans. Among
-
for 12.4% of deaths in this demographic.⏎While opioid-related deaths were common among older age groups as well — there were about 6,700 among
-
in the overweight range is associated with increased risk of poor health for younger adults, but the opposite is true in older age. A number of studies of people in their sixties and beyond have shown that
-
AMA Network Open, reservation-based American Indian students are at a higher risk for substance use compared with US youth in general, highlighting the necessity of increased intervention efforts for this population.⏎Researchers conducted a population-based survey study of 8th, 1
-
), examines a paradox in African American drinking. African Americans report initiation to drinking at an older age, lower rates of use and lower levels of use in nearly all age groups. Nonetheless, the group encounters higher levels of problems related to alcohol when compared
-
2018, about 21.2 million people age 12 and older, or approximately 1 in 13 people in this age group, needed substance abuse treatment.1 People go untreated for many reasons. They
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 have a 3D shape in 3D space, with both natural and natural natural natural natural natural natural natural natural natural natural natural natural natural natural natural natural natural natural natural natural natural natural natural natural natural natural natural natural natural natural natural
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
) is the same.⏎- To this day we follow the model that Galileo created.⏎ag = g = 9.81m/s2⏎ag = g = acceleration due to gravity⏎Since
-
fun facts about America's first President of the United States, from this height and weight to his favorite foods.⏎Grades: PreK–K, 1–2⏎In this lesson, students read about George Washington and discuss the monument built
-
latest in professional publications, learn new techniques and strategies, and find out how you can connect with other literacy professionals.⏎Teacher Resources by Grade⏎|1st - 2nd||3rd - 4th|⏎|5th - 6
-
' Classroom.|⏎|Levels:||Late Elem. (3-5), Middle School (6-8), High School (9-12
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
- L21:1964 excites
- L19:2633 excites
- L20:1254 suppresses
- L16:2246 weakly excites
- L12:2415 weakly suppresses
- L21:2659 weakly excites
- L3:2030 weakly suppresses 'home' after 'the' or 'in'
- L6:2704 weakly excites
- L9:3023 weakly suppresses
- L22:1196 weakly excites
- L21:1744 weakly suppresses
- L11:2980 weakly excites
- L17:2528 weakly suppresses
- L19:2289 weakly suppresses
- L5:2595 weakly excites the word 'human' in science text
- L4:1099 weakly excites
- L22:1080 weakly suppresses
- L12:466 weakly suppresses
- L22:1803 weakly excites
- L11:2738 weakly excites
- L9:1855 weakly suppresses
- L22:2133 weakly excites
- L21:1162 weakly excites
- L4:107 weakly suppresses
Feeds later layers
- L27:1257 strongly excites
- L24:1208 strongly excites
- L26:2059 excites
- L25:3048 excites
- L25:340 excites
- L25:2598 weakly suppresses
- L24:687 weakly excites
- L25:2827 weakly excites
- L24:737 weakly suppresses
- L24:2619 weakly suppresses
- L25:2464 weakly excites
- L25:587 weakly suppresses
- L26:481 weakly suppresses
- L26:2024 weakly excites
- L26:1728 weakly excites
- L24:1353 weakly excites
- L26:2410 weakly excites
- L24:827 weakly excites
- L24:776 weakly suppresses
- L24:445 weakly suppresses
- L25:2360 weakly excites
- L27:679 weakly suppresses
- L26:2332 weakly excites
- L25:270 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
- group
- groups
- group
- groups
- Group
- Groups
- Group
- _group
- -group
- (group
Suppresses
- mland
- <TreeNode
- PAD
- hlen
- boxing
- xaa
- ScreenState
- ialias
- resourceName
- nạn