page numbers in citations
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. 14 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 responds to the digits of page numbers in academic references, such as the '3' in '14(3)' or the blank after 'pp.' where a page number should come. Turning it up gave unrelated text, but the strongest examples are all citations.
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.
-
<|endoftext|>Philosophy East and West 46 (2):143-164 (1996)⏎|Abstract||The origin,
-
.⏎Katrin Froese (2008). Organic Virtue: Reading Mencius with Rousseau. Asian Philosophy 18 (1):83 – 104.⏎Tao
-
L. Nilsson-Stutz (Eds.), The Oxford Handbook of the Archaeology of Death and Burial (Oxford University Press, Oxford), 631–642, 2013. My thoughts precipitated
-
with Mencius. Asian Philosophy 9 (2):123 – 133.⏎Qingping Liu (2001).
-
.B., & Lapp, D. (Eds.), Handbook of research on teaching literacy through the communicative and visual arts (Vol 2, pp. 4
-
and the persuasiveness of viral campaigns in social network sites: Persuasive intent as the underlying mechanism. Journal of Marketing, 18(1), 39-53. http://dx.doi.org/10.
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 been a student for a long time, and I have taken a trip for a short time, and I have gone through a period of time where I had been in my own life for a long time, and I had gone through
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
and rheumatic fever in children: time trend analysis. BMJ. August 6, 2005;331:328–9.⏎Used with permission from
-
and Freedom”, Nature, Nov. 15, 1941, Vol. 148, pp. 580-584.⏎The reasons why the adoption
-
people within a matter of seconds. See George C. Wilson, Visible Violence, 12 NAT’L J. 886, 887 (2003). Short of bombs, missiles, and biochemical agents,
-
ate the molecular basis of the genetic susceptibility to alcoholism.(JAMA. 1992;268:1877-1882)<|endoftext|>The Imaginative Treatment of Geometry in Waldorf Education⏎A
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
- L11:921 weakly suppresses
- L8:2030 weakly excites
- L9:2364 weakly suppresses
- L10:1666 weakly excites
- L10:2463 weakly suppresses
- L13:2238 weakly excites
- L13:1216 weakly excites
- L6:2891 weakly suppresses
- L5:388 weakly excites number ranges like 0 to 9
- L12:738 weakly excites
- L2:1884 weakly suppresses
- L10:2879 weakly suppresses
- L5:2753 weakly excites
- L12:124 weakly excites
- L10:1434 weakly suppresses
- L11:36 weakly excites
- L11:1267 weakly suppresses
- L13:2365 weakly suppresses
- L5:133 weakly excites
- L8:561 weakly suppresses
- L0:193 weakly excites
- L11:1443 weakly excites
- L11:2741 weakly suppresses
- L1:1481 weakly excites
Feeds later layers
- L21:676 excites
- L18:585 weakly excites
- L17:2513 weakly excites
- L15:1279 weakly suppresses
- L18:2700 weakly suppresses
- L17:2082 weakly excites
- L22:1684 weakly excites
- L20:1709 weakly suppresses
- L25:2915 weakly suppresses
- L19:837 weakly suppresses
- L18:2534 weakly suppresses
- L22:2702 weakly excites
- L25:2370 weakly suppresses
- L20:2433 weakly suppresses
- L22:2772 weakly suppresses
- L21:1723 weakly excites
- L18:658 weakly excites
- L15:2649 weakly suppresses
- L27:1012 weakly suppresses
- L16:1877 weakly suppresses
- L18:2644 weakly suppresses
- L19:1222 weakly excites
- L27:1138 weakly suppresses
- L17:523 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
- 率达到
- $page
- 页
- tabpanel
- (page
- →page
- page
- ?page
- ipp
- อบ
Suppresses
- ought
- engl
- loomberg
- umph
- ɐ
- acional
- ernal
- elden
- qua
- ನ