comma after 'As a [role]'
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. 5 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 the comma right after phrases like 'As humans,' or 'As physicians,' that introduce a 'we' statement. Turning it up made the model write in a more general, reflective voice.
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.
-
years ago. We are the only leaf remaining on our twig, and we are dependent on the same tree we share with millions of other leaves…⏎As humankind, we are all part of the same leaf…⏎The differences we have seen in each
-
. Fresh or fermented cocoa beans contain about 10% flavan<|endoftext|>The eyes are the windows to the soul, or so goes the proverb. As humans, we use our eyes to express a wide variety of emotions, often without even realizing
-
=EE workgroup has been meeting for several months and plans focus groups of health professionals and community members.⏎"We're all responsible," Lucas said. "As individuals, as patients, we have to be more responsible for our own health. We have
-
of the state, they have enormous authority."(2) Richard H. Minear, Japanese historian, answers the question this way: "As a practicing historian, I encounter at every turn the power textbooks exercise over my students' minds. In
-
floss more? Studies show that 7 out of 10 Canadians will develop gum disease at some point in their lifetime. 1 As hygienists, it is our goal (if not our dream
-
Fam Pract Manag. 2013;20(5):40⏎Author disclosure: no relevant financial affiliations disclosed.⏎As physicians, we often like to think that the doctor- patient relationship is the key to effective
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 user is not familiar with in the current context of the research being conducted. This is a technical question that is related to the user's interest. This is a technical question that is related to the user's interest. This is a technical
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
and draughting – were under scrutiny too, for the scientific and artistic expertise involved in making maps in the past was integral to communicating through cartography, as indeed it still is today. The colloquium programme and abstracts are available here
-
in a puritan society. In his novel,The Scarlet Letter, Nath<|endoftext|>What they learned after only seven years is worth remembering (or learning, as the case may be).⏎The United States has operated under two constitutions. The first,
-
of that time, as I did—was that at least one of the editors, Malcolm Bird, was not so
-
is a complex, global issue that threatens the sustainability and resilience of our fishing communities,<|endoftext|>Suicide and Social Change in China⏎- Cite this article as:⏎- Phillips, M.R., Liu, H. & Zhang, Y. Cult
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
- L2:1880 weakly suppresses
- L4:2509 weakly suppresses
- L4:2539 weakly suppresses
- L4:1441 weakly excites
- L2:530 weakly suppresses
- L2:1843 weakly excites
- L2:1254 weakly excites
- L3:2024 weakly suppresses
- L1:295 weakly suppresses
- L4:451 weakly suppresses
- L1:472 weakly suppresses
- L4:2260 weakly excites
- L1:2009 weakly excites
- L4:2805 weakly excites
- L3:961 weakly excites
- L3:2028 weakly excites
- L1:1235 weakly excites
- L0:1178 weakly suppresses
- L1:1776 weakly suppresses
- L4:2711 weakly suppresses
- L3:249 weakly excites
- L0:386 weakly excites
- L3:1010 weakly suppresses "to" in "click here to"
- L3:115 weakly suppresses
Feeds later layers
- L6:2283 excites
- L9:2442 excites
- L6:1001 weakly excites
- L11:1072 weakly excites
- L12:1295 weakly excites
- L17:2717 weakly excites
- L11:949 weakly excites
- L13:2736 weakly excites
- L22:1457 weakly suppresses
- L7:1746 weakly excites
- L24:2408 weakly suppresses
- L11:1786 weakly excites
- L16:954 weakly suppresses
- L23:2703 weakly suppresses
- L14:3044 weakly excites
- L25:2902 weakly suppresses
- L15:1129 weakly suppresses
- L20:813 weakly excites
- L11:2135 weakly suppresses
- L10:371 weakly excites
- L27:816 weakly suppresses
- L17:1845 weakly excites
- L11:2104 weakly suppresses
- L8:611 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
- 也不能
- 只能说
- ọng
- 也只能
- 看
- 看他
- 看电影
- 只能
- ModelState
- &view
Suppresses
- wig
- حول
- 坞
- /O
- BarController
- .getHeader
- ólogo
- corridors
- Ahmad
- ueur