names and titles inside proper nouns
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. 16 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 middle of a fixed name or title, such as 'Age' in 'An Age', 'Nation' in 'A Nation at Risk', or 'End' in 'The End of the Cold War'. It is quiet, or pushed negative, on phrases like 'Call Before You Dig' and 'All Others Are Silent'. Turning it up made the output repeat itself.
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.
-
. S. cabinet departments, including those overseeing commerce, education, energy, and labor (some of this material and references for it can be found in my An Age
-
approach to what is called "hands-on" science has been growing in recent years at all levels of education, since the 1983 report "A Nation at Risk" criticized public education, including science curriculums, in American
-
still honor the format of exploring hands-on activities, learning about artwork in our galleries, and making an inspired art project.⏎As always, Kid Quest is an at-your-own-pace art adventure. In this case, take all the time you
-
This video is part two of a seven-part National Academies series, Climate Change: Lines of Evidence. The video outlines, with the use of recent research and historical
-
68" onward, had little effect but to strengthen comparable hard-line views in the Kremlin.⏎Armstrong, David, and Erik Goldstein, eds. The End of the Cold War. Routledge, 2013.⏎
-
), Part 4, Chapter 7, eBook.<|endoftext|>Transcript for A New Horizon, segment 05 of 13⏎The Eastern Colorado Project's
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, as... and I have been... and I have been... and I have been... and I have been... and I have been... and I have been... and I have been... and I have been... and I
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
in the upper echelons. Richard would have probably received these treatments because he was a member of the nobility."<|endoftext|>When Leaders Speak… All Others Are Silent!⏎December 22, 2015 | 3 min read
-
From Tree Roots⏎- Locate sewer lines – You can call your local public works office or the national 811 “Call Before You Dig” number to locate underground utilities on your property. Before you do any kind of
-
. There were numerous reports of NYC residents unable to get their phone calls through on 9/11, because the system was jammed.⏎Business Before Pleasure on the Wire: Effect of ‘Phone Philandering on the Call Frequency Curve of
-
ving⏎Tier 2—Assessment and Problem Solving⏎We can no longer implement instruction we believe to be great and cross our fingers and hope for the best. We must identify appropriate strategies the way a physician determines an appropriate prescription. Then
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
- L14:3049 suppresses
- L15:1395 suppresses
- L13:1531 suppresses
- L13:1550 suppresses
- L12:2126 suppresses
- L13:2938 excites
- L14:1631 excites
- L14:282 suppresses
- L7:2596 suppresses
- L8:1267 excites
- L15:2459 suppresses
- L7:2155 suppresses
- L15:1191 suppresses
- L4:881 excites
- L13:2983 weakly suppresses
- L15:354 weakly excites
- L12:1831 weakly excites
- L13:1742 weakly excites
- L4:1026 weakly suppresses
- L13:1945 weakly excites
- L15:484 weakly excites
- L11:2815 weakly suppresses
- L13:293 weakly excites
- L7:1356 weakly suppresses
Feeds later layers
- L17:76 excites
- L17:375 suppresses
- L20:353 suppresses
- L17:293 suppresses
- L17:112 suppresses
- L25:334 suppresses
- L24:121 suppresses
- L17:186 suppresses
- L20:320 suppresses
- L24:78 excites
- L17:181 excites
- L19:211 excites
- L19:1201 suppresses
- L17:48 suppresses
- L20:202 suppresses
- L20:218 suppresses
- L18:160 excites
- L20:34 excites
- L17:937 suppresses
- L18:1120 suppresses
- L23:40 suppresses
- L17:2075 suppresses
- L21:365 excites
- L18:360 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
- ivial
- createAction
- …⏎⏎
- ixin
- omat
- omon
- 内外
- ون
- もの
- ait
Suppresses
- Slovak
- 즐
- 院院士
- !
- 佶
- iced
- .bill
- angelo
- ยาย
- _currency