'of' after a noun
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. 1 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 the word 'of' when it follows a noun such as 'designs', 'contents' or 'modifications'. When it was turned up, the model's text did not visibly change in a way that matched this pattern, so the link to the output is unclear.
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.
-
⏎This compendium comprises articles on legal and technological issues of Internet Governance in sphere of human rights prepared by the group of legal and technical scholars of information studies of the Higher School of Economics. This compound is devoted to the workshop on ‘Human
-
especially when the hunting pressure increases.⏎Last season I spent some time examining stomach contents of harvested deer in the pine woods of East Feliciana Parish and in the bottom
-
in the early 1930s. But that didn’t stop companies from speculating about TV designs of the future. Like this Marconi-brand “radio set of 196
-
ERS.⏎The post 1929 Stock Market Crash and the Great Depression. first appeared on essaypanel.com.<|endoftext|>This tutorial shows how to send modifications of code in the right way: by using patches.⏎
-
is identified as a counsel, usually with a flat area, of a complete or component of a region. The job of any map is usually to describe spatial interactions of distinct characteristics that this map aspires to signify. There are several forms of maps
-
deepening sense of self and trust in their ability to stand up for themselves, with peers and beyond.⏎This approach to teaching them to argue—through our behaviors of patience, reconsidering, acknowledging, and welcoming their influence in our thinking—takes
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 thinking about something. I have been thinking about something. I have been thinking about something. I have been thinking about something. I have been thinking about something. I have been thinking about something. I have been thinking about something
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
ty to Animals (ASPCA), "Animals in Entertainment" refers to any animal(s) used to act, perform, fight and/or kill for the enjoyment of humans. The term encompasses many different forms of entertainment – from circuses to movies
-
eyes.<|endoftext|>In January, CLBB hosted a panel discussion, “Empathy: The Development and Disintegration of Human Connection,” meant to explore the remarkable capacity of humans to relate to others,
-
America placed first, second or third in 20 events while Greece scored in 46 events and Germany placed in 13 competitions. To the delight of the hometown crowd, Greek runner Spyridon Louis won the marathon. The first
-
treaties ‘recall that human rights are universal and shall apply to all individuals, and stressing therefore its commitment to guarantee the equal dignity of all human beings and the enjoyment of rights and freedoms of all individuals without discrimination on any ground’ .However, despite
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
- L0:1257 weakly suppresses
- L0:1525 weakly excites
- L0:1531 weakly excites
- L0:1220 weakly suppresses
- L0:277 weakly excites
- L0:15 weakly excites
- L0:1219 weakly suppresses
- L0:405 weakly suppresses
- L0:2846 weakly excites
- L0:178 weakly suppresses
- L0:1887 weakly suppresses
- L0:1604 weakly excites
- L0:2183 weakly excites
- L0:2547 weakly excites
- L0:2300 weakly excites
- L0:1224 weakly suppresses
- L0:120 weakly excites
- L0:510 weakly suppresses
- L0:451 weakly suppresses
- L0:2917 weakly suppresses
- L0:2528 weakly suppresses
- L0:919 weakly suppresses
- L0:968 weakly excites "and" joining two nouns
- L0:1123 weakly suppresses
Feeds later layers
- L2:1837 excites
- L7:416 weakly excites
- L20:2141 weakly suppresses
- L2:1124 weakly suppresses
- L16:1089 weakly suppresses
- L22:1657 weakly suppresses
- L7:243 weakly excites
- L5:546 weakly excites
- L3:1187 weakly suppresses
- L26:37 weakly excites
- L17:289 weakly excites
- L19:988 weakly suppresses
- L18:2883 weakly suppresses
- L16:1133 weakly suppresses
- L17:2353 weakly suppresses
- L16:2997 weakly suppresses
- L22:470 weakly suppresses
- L15:649 weakly suppresses
- L5:630 weakly excites
- L7:2291 weakly suppresses
- L20:887 weakly suppresses
- L9:2921 weakly suppresses
- L20:2121 weakly suppresses
- L16:1098 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
- arges
- lore
- Fon
- ARGE
- Rendering
- leet
- .Fill
- conna
- وع
- ASTE
Suppresses
- //{⏎
- rowable
- 那就是
- pire
- ishop
- серии
- 的就是
- 因为她
- iteit
- 但她