'World' in organisation names
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. 25 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 'World' in names like World Health Organization, World Factbook and World Heritage, and also on a few 'are' and 'was' words after a definition. Turning it up made the text talk about things to consider and the world in general, while turning it down gave loose, rambling text.
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.
-
locate.<|endoftext|>Science Fair Project Encyclopedia⏎The World Factbook⏎The World Factbook is an annual publication by the Central Intelligence Agency
-
PM2.5, mainly from coal.⏎As it stands now, only about two percent of China’s nearly 1.4 billion people breathe air that meets World Health Organization (WHO) guidelines, the IEA said.⏎That proportion would rise
-
in Poland in 1473 and as a young man became fascinated by math, medicine and science, particularly astronomy. His studies led him to challenge what was then a widely accepted belief — that the Earth stood still in outer space while the
-
of brucellosis was considered confirmed if results of serologic tests, such as ELISA or complement fixation test, were positive, in accordance with the World Health Organization case definition (<|endoftext|>The
-
While earlier versions of the laws dealt with cattle and livestock, today Texas cruelty laws protect a variety of animals from inhumane treatment. These laws define which actions are “cruel”, explain what punishment or penalties can be imposed against those whom
-
or neither?⏎Parallel lines are the lines that never intersect and perpendicular lines are the lines that intersect at 90 degrees.⏎Answer: Lines with the same slope are parallel and if the slope of one line is the negative reciprocal of the second line
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 outting out out of the best of my best outing outing outing outing outing outing outing outing outing outing outing outing outing outing outing outing outing out
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
of the monument. Where it is not available, the satellite view is shown instead.⏎Latitude: 52.5162 / 52°30'58"N⏎Longitude: -3.0604 /
-
UNESCO World Heritage Committee has put the Site on its List for the following reasons:⏎The realization of an architectural and sacred art’s work into a natural landscape for didactic and religious aims, finds its⏎highest expression in northern Italy Sacred Mounts
-
individually, and asked the confederate to leave for a few minutes. After the confederate had left, the experimenter debrief<|endoftext|>Passchendaele at Home⏎Passchendaele at Home was a nationwide initiative
-
understanding of his culture and the injustices Navajos faced. Bruchac sprinkles historical facts throughout the book in a way that feels fascinating, rather than didactic. I learned so much from Code Talker. My high school boyfriend was
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
- L22:2721 strongly excites
- L20:2988 excites
- L15:1531 excites
- L18:2451 suppresses
- L19:2286 suppresses
- L11:1484 weakly suppresses
- L9:66 weakly excites
- L0:1609 weakly suppresses
- L21:2482 weakly suppresses
- L16:2661 weakly suppresses
- L20:2406 weakly suppresses
- L10:2934 weakly excites
- L22:2071 weakly excites
- L8:1219 weakly suppresses
- L19:1612 weakly excites
- L19:386 weakly suppresses
- L19:2748 weakly suppresses
- L7:1265 weakly suppresses
- L17:2645 weakly suppresses
- L15:599 weakly excites
- L20:1872 weakly suppresses
- L19:1884 weakly suppresses
- L19:2495 weakly suppresses
- L20:2121 weakly suppresses
Feeds later layers
- L27:1849 strongly excites
- L27:834 weakly excites
- L27:879 weakly excites
- L27:526 weakly excites
- L27:310 weakly suppresses
- L27:2319 weakly suppresses
- L26:2023 weakly suppresses
- L26:612 weakly suppresses end-of-text boundary tokens
- L27:2291 weakly suppresses
- L26:2540 weakly excites
- L27:431 weakly excites
- L27:1282 weakly excites
- L26:2369 weakly suppresses
- L27:2677 weakly suppresses
- L27:2829 weakly excites
- L26:1688 weakly excites
- L26:1349 weakly excites
- L27:1049 weakly suppresses
- L27:855 weakly suppresses
- L26:1877 weakly suppresses
- L27:997 weakly suppresses
- L26:2383 weakly excites
- L27:1399 weakly suppresses
- L26:1585 weakly 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
- consider
- considered
- 考虑
- Consider
- consideration
- considers
- Consider
- consider
- consid
- considerations
Suppresses
- pel
- ault
- 确定
- UTES
- 定
- emat
- 现有的
- "path
- ать
- amework