commas inside lists of items
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. 3 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 a comma in the middle of a list of things, typically just before the last item or "and". Turning it up did not produce any clear theme in the generated text.
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.
-
. Some of the activities include making a jester hat, learning the chivalric code, making a shield, making a trebuchet, and more.⏎- Knowledge Box Central families: Use this alongside the Middle Ages lap
-
fine motor skills to cut out small numbers, think about how to solve six types of math problems, judge the difficulty level and their own capabilities, develop spatial awareness, and learn to apply the correct amount of glue (another fine motor skill, and
-
Voyant Tools: Voyant Tools is a free web based suite of tools that allows for you to copy and paste text into a textbox, upload a file, or use an already predetermined corpus and use different tools to visualize your text. The
-
Teach them that a good person chooses to think good thoughts, not to say profaned words, do good deeds, go to good places, hang with good people, and find reasons not to fight. Remember your actions speak louder than your words.⏎
-
of interest to you.⏎In this guide, we’ll go over all of this in detail. We’ll explore the different types of cobots, their benefits, and how they compare to industrial robots so that you can become better acquainted with this
-
Keywords: viral marketing, word-of-mouth marketing, consumption patterns, information traits, credibility.⏎ANTHEUNIS, M.L.,
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 first person in the room is the first person in the room, the first person in the room, the first person in the room, the first person in the room, the first person in the room, the first person in the room
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
HGUOT ; reverse order⏎mov char, al⏎L1: mov eax,32 ; creates ASCII code decrement by 32⏎mov char, AL⏎mov al 5-character⏎mov al, TYPE BYTE ; 6
-
populations changed over time.⏎Further studies involving more infants could eventually help researchers understand the causes of various intestinal problems that afflict preemies, in particular the sometimes fatal necrotizing enterocolitis, according to researchers at the University of California, Berkeley,
-
months in the intensive care unit before they're deemed healthy enough to go home. Between 5 and 10 percent of these preemies develop symptoms of necrotizing enterocolitis (NEC), which requires rounds of antibiotics to halt,
-
"⏎- Brix Orolfo, age 12yrs.⏎- © Brix Orolfo⏎Winner
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
- L2:66 excites
- L2:284 suppresses
- L0:2921 suppresses
- L2:527 weakly excites
- L1:86 weakly excites
- L2:36 weakly excites
- L0:160 weakly suppresses
- L0:2838 weakly excites
- L0:1596 weakly excites
- L2:2202 weakly excites
- L2:2784 weakly suppresses
- L0:1756 weakly suppresses
- L0:1030 weakly excites
- L2:1695 weakly excites
- L0:178 weakly suppresses
- L2:49 weakly suppresses
- L2:1879 weakly suppresses
- L1:1029 weakly excites
- L0:1119 weakly suppresses
- L2:75 weakly suppresses
- L1:1975 weakly suppresses
- L0:2007 weakly excites 'key' after 'The'
- L0:2247 weakly suppresses
- L0:1511 weakly suppresses
Feeds later layers
- L22:681 excites
- L19:2732 weakly suppresses
- L8:454 weakly excites dash starting a new list item
- L11:2594 weakly suppresses
- L9:2666 weakly suppresses
- L5:1019 weakly suppresses
- L15:2268 weakly suppresses
- L4:44 weakly excites
- L14:39 weakly suppresses
- L20:1752 weakly excites
- L8:1460 weakly suppresses
- L4:281 weakly suppresses
- L17:1287 weakly suppresses
- L7:237 weakly excites
- L7:2233 weakly excites
- L25:524 weakly suppresses
- L11:1468 weakly suppresses
- L5:2400 weakly suppresses
- L14:389 weakly suppresses
- L26:106 weakly suppresses
- L9:1248 weakly suppresses
- L4:96 weakly suppresses
- L15:2677 weakly suppresses
- L7:1055 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
- engers
- 惮
- yd
- erokee
- heel
- ilarity
- .default
- 发展格局
- ILES
- ulum
Suppresses
- .Access
- .ACCESS
- 手
- .access
- tain
- เหร
- 本报
- themselves
- .setColumns
- ammonia