"over" in "all over the world"
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 lights up on the word "over" when it comes right after "all", as in "all over the world". Turning it up made the model's text drift into unrelated topics like code and random puzzles, so the steering did not show a clear theme.
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.
-
and distributed to classrooms all over the world. They also offer workshops to assist K-12 teachers interested in
-
jet stream and producing more ferocious hurricanes and floods all over the world. But its effects will be acutely felt here, according to a
-
s" in some of the oddest of places such as the tops of power line poles. They have been seen in Chicago, New York as well as all over South Florida. Speculation as to how these feral flocks developed seems topping
-
video is not just about having fun, it brings people together from all over the world.⏎With the right music, it now becomes an inspirational video that can
-
, engineers will be able to construct microthin digital displays that can be<|endoftext|>Diabetes mellitus is a non communicable disease but is spreading like a pandemic all over the world. Each and every house is having at least two members who are diagnosed
-
Patch, consists of waste such as toilet seats, camera cases, and oil drums, floating in the middle of the Pacific ocean. The trash accumulates from all over the world and as it moves into the ocean, it becomes trapped at the center
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: "In a 2 game, the two players are A and B. If A has a strategy of 1, what should the other player do?" (A) and (B) to respond? The other player has a strategy of
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
R. L., Sobel, R. S., & Macpherson, D. A. (2018). Economics: Private⏎and public choice (16
-
be found professionals and graduates of British and American universities.⏎A few trade unions had existed for the protection of workers of certain categories (e.g., the Nigeria Civil Service Union) before trade unionism became legalised in 1938.
-
makes it easier for<|endoftext|>The latest weapon unleashed to battle California's growing energy demand comes in the form of free software. The State of California Energy Commission's Public Interest
-
also read the full article published in Transactions of the Institute of British Geographers here.⏎Ethnic diversity in English city neighbourhoods has been the subject of longstanding public interest. While some have argued that diversity is associated with lower social cohesion and trust,
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
- L0:1301 suppresses
- L0:1224 suppresses
- L0:178 suppresses
- L0:1 weakly suppresses
- L0:1229 weakly excites
- L0:512 weakly suppresses
- L0:179 weakly suppresses
- L0:571 weakly excites
- L0:1638 weakly excites
- L0:2351 weakly excites
- L0:732 weakly excites
- L0:2436 weakly suppresses
- L0:1660 weakly excites
- L0:2408 weakly excites
- L0:2645 weakly excites
- L0:1167 weakly suppresses
- L0:2089 weakly suppresses
- L0:943 weakly excites
- L0:12 weakly suppresses
- L0:187 weakly excites
- L0:1554 weakly excites
- L0:2522 weakly excites
- L0:1075 weakly excites
- L0:209 weakly excites
Feeds later layers
- L13:2627 weakly suppresses
- L8:2160 weakly excites
- L21:1547 weakly suppresses
- L9:2479 weakly suppresses
- L3:411 weakly suppresses
- L12:2502 weakly suppresses
- L4:2276 weakly suppresses
- L8:3065 weakly suppresses
- L2:2399 weakly suppresses
- L25:400 weakly suppresses
- L22:1673 weakly suppresses
- L7:1926 weakly suppresses
- L27:2658 weakly suppresses
- L21:2097 weakly excites
- L26:2117 weakly suppresses
- L13:3067 weakly suppresses
- L2:1868 weakly excites
- L2:2591 weakly suppresses
- L3:2420 weakly excites
- L11:2267 weakly suppresses
- L2:2726 weakly excites 'by' after 'characterized' or 'marked'
- L4:2456 weakly suppresses
- L9:2810 weakly excites
- L5:1691 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
- erable
- ически
- ome
- omes
- lobal
- Respect
- ическое
- ейств
- ened
- ergy
Suppresses
- bull
- -zA
- /pub
- >*</
- |_|
- 专区
- >'+⏎
- _('
- mutation
- $/,⏎