unclear: book and report titles
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. 17 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. 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 words inside titles and descriptions of books, reports and academic topics, but the pattern is scattered and it also fires weakly on random number and symbol fragments. Turning it up made the model drift into citation-like, academic-sounding text, though the other outputs were mostly unrelated.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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.
-
and 1930s, a time when avant-garde movements transformed writers' and readers' conceptions of literature. Urban Chroniclers in Modern Latin America: The Shared Intimacy of Everyday
-
in his country and intentionally gave the refugees who passed through Portugal during the war a hard—but not impossible—time.⏎In Hitler’s Jewish Refugees: Hope and Anxiety in Portugal, the distinguished historian of modern German Jewry Marion Kaplan is less interested
-
, please check our FAQs, and if you can't find the answer there, please contact us.<|endoftext|>Indigenous Peoples and Climate Change: from victims to change agents through decent work⏎The present report analyses the situation of indigenous peoples in the context
-
, often without organized funding or direct support from the church hierarchy. A comprehensive history of Roman Catholic nuns and sisters in the American West, Across God's Frontiers reveals Catholic sisters as dynamic and creative architects of civic and religious institutions in western communities
-
and Modern Ireland, 1782-1992 (1994).⏎Chapter 1 The Easter Rising 1⏎Chapter
-
s and now seeing renewed interest, uses liquid fuel made of fluoride thorium and a salt. There are safety advantages<|endoftext|>The Path from Discontentment to Contentment⏎Shakespeare’s play Richard III, Act-I, Scene-I, Lines
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 idea of, but it's not, the idea of, but it's not, the idea of, but it's not, the idea of, but it's not, the idea of, but it's not, the idea of
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
04) 13: 2511. doi:10.1023/B:BIOC.0000048451.07820.9c⏎- 1
-
://mynursingpaper.com/wp-content/uploads/2020/06/cropped-155972031988413967-1-768x146
-
pp. 125–135. doi:10.1112/S0025579300008718.<|endoftext|>The Air We Breathe: Are Newer
-
mynursingpaper.com/wp-content/uploads/2020/06/cropped-155972031988413967-1-768x146
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
- L13:1325 excites
- L14:1909 excites
- L15:1066 suppresses
- L15:2532 suppresses
- L10:270 suppresses
- L12:996 suppresses
- L16:2398 excites
- L13:852 excites
- L12:2494 suppresses
- L14:895 excites
- L13:1632 suppresses words inside headlines and titles
- L14:790 suppresses
- L14:2061 excites
- L16:2471 suppresses
- L12:1956 suppresses
- L16:1047 excites
- L15:2591 excites
- L15:2459 suppresses
- L16:590 excites
- L7:1656 excites
- L13:1797 suppresses
- L15:655 suppresses
- L6:234 suppresses
- L12:2654 excites
Feeds later layers
- L22:1712 excites colon before a subtitle
- L18:830 excites opening quote marks before a coined term
- L18:2412 excites
- L18:2706 excites
- L19:2461 excites
- L19:260 excites
- L23:526 excites
- L21:1078 excites
- L21:1067 excites
- L18:100 weakly excites
- L27:1103 weakly excites
- L21:2216 weakly excites
- L19:2780 weakly excites
- L19:2458 weakly excites
- L18:2611 weakly excites
- L20:36 weakly suppresses
- L19:950 weakly excites
- L18:2933 weakly excites
- L18:113 weakly excites
- L26:1721 weakly excites
- L21:2020 weakly excites
- L20:758 weakly excites
- L19:1896 weakly excites
- L18:1287 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
- -part
- 部分
- PART
- ernel
- 》,
- SEMB
- 第九
- (parts
- ichte
- Part
Suppresses
- …"⏎⏎
- ..."⏎
- 啬
- )">⏎
- ;"⏎
- ..."⏎⏎
- clap
- ...">⏎
- urat
- =num