'of' in organisation names like 'Ministry of'
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. 5 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 'of' inside the names of government bodies and institutions, like 'Ministry of Health' or 'School of Music'. Turning it up didn't produce a clear theme in the sample 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.
-
1921-School of Music organized.⏎1923-School of Business organized.⏎195
-
year, WAVES program stated that “African countries lead on natural capital accounting”. The recent flurry of activity on the continent appears to support this.⏎The Ministry of Finance and WAVES organized a training as part of ongoing efforts to implement NCA
-
2,200 people, according to the World Health Organization. Because a dead body can transmit the virus for up to a week after death, the Ministry of Health prohibited all traditional burials,
-
2021 and 2022, respectively.⏎This speed is far below the speed required to achieve the US decarbonization target. The Ministry of Energy’s solar energy research shows that the new installed capacity in the next five years
-
a health facility without a skilled birth provider. Consequently, teaching basic newborn care along with other basic health skills to community care providers is a vital part of the Ministry of Public Health's national strategy.⏎Together, USAID and the Afghan Ministry of Health
-
large number of local manifestations of the element are inscribed on the National List of Intangible Cultural Heritage administered by the Department of Intangible Cultural Heritage of the Ministry of Culture.⏎These videos (and many more) can also be consulted through the UNESCO
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: it's not a breaking score. It's a bad one. So, it's not a bad one. So, it's a bad one. So, it's a bad one. So, it's a bad one. So,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
night to make it to Willamette’s 1929 freshman orientation. A month later the stock market collapsed
-
tutorial from the Australia Telescope National Facility, Parkes, New South Wales⏎Amateurfunk, Erde-Mond-Erde, EME
-
and we have the power to stop it.”⏎The event is sponsored by the Jewish Student Union and supported by Campus Ambassadors, Students Against Slavery and Willamette University's Religious Studies Department.⏎For more information contact Jessica Sternbach at firstname.lastname
-
is approved, endorsed, or authorized by the Central Intelligence Agency. "⏎Many sites have used information and images from the CIA World Factbook, because of its public domain status, including this encyclopedia.⏎Besides the World Factbook, the CIA puts out
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
- L4:977 suppresses
- L0:1695 weakly suppresses
- L1:298 weakly suppresses
- L0:2278 weakly suppresses
- L4:1588 weakly suppresses
- L1:506 weakly excites
- L1:1414 weakly excites
- L4:876 weakly excites
- L3:1950 weakly excites
- L1:2217 weakly suppresses unclear: mostly weak, Chinese and Greek characters
- L3:1033 weakly excites
- L2:1306 weakly suppresses
- L2:2500 weakly excites
- L4:2588 weakly excites
- L0:271 weakly suppresses
- L3:2053 weakly suppresses
- L4:979 weakly suppresses
- L2:420 weakly suppresses
- L3:554 weakly excites
- L3:456 weakly suppresses
- L2:2805 weakly suppresses
- L4:92 weakly suppresses
- L4:2825 weakly suppresses "to" after 'for X' infinitives
- L2:2545 weakly excites
Feeds later layers
- L6:554 weakly suppresses
- L7:581 weakly excites
- L13:236 weakly excites
- L7:285 weakly excites
- L24:2273 weakly excites
- L6:201 weakly suppresses
- L15:723 weakly suppresses
- L8:1682 weakly excites
- L16:1992 weakly excites
- L7:1403 weakly excites 'to' after attempt or seek verbs
- L6:1539 weakly excites
- L23:2857 weakly suppresses
- L21:1197 weakly excites
- L8:1946 weakly excites comma after 'who' or 'which'
- L18:3060 weakly excites
- L15:1039 weakly excites
- L9:80 weakly excites
- L22:1621 weakly excites
- L6:47 weakly suppresses
- L9:817 weakly excites
- L14:1474 weakly suppresses
- L9:868 weakly suppresses
- L13:270 weakly excites
- L14:208 weakly excites 'of' or 'the' after 'part of' or 'subset of'
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
- getApplication
- omon
- currentItem
- aci
- amation
- ment
- emme
- IGATION
- informação
- 豉
Suppresses
- eer
- hdl
- веч
- asl
- iful
- ush
- cosity
- ensored
- coon
- олос