'and' starting a place or person name
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. 22 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 letters 'And' when they start a place or name, such as Andes, Andaman, Andhra or Andromeda. Turning it up made the model repeat the word 'yet', which the neuron does not clearly relate to its trigger.
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.
-
. So, for example, China with its population of 1.3 billion people gets 1,900 electoral-college votes, but even tiny Andorra with less than 500,000 people gets three electoral
-
their reflections off the serene lake. Visible in the distant background sky is the band of our Milky Way Galaxy, the Pleiades open clusters of stars and the Andromeda galaxy.”⏎A NASA original, this galaxy is appropriately named for it’s
-
the geology department, gave an account of the activities of his department.⏎Later, Dr. Bhushan told reporters that after the Sumatra earthquake, the Andaman and Nicobar islands experienced 200 aftershocks of the magnitude
-
ta⏎called a jubba. Both of these clothes are called "kurtas." In the state⏎of Andhra Pradesh, both of these options are common. Either of these shirts⏎would
-
emphasis on downwind new fires occuring.⏎The highest spots are in south central, north central and northeast South Dakota. Just this week, the area near Lake Andes is also considered at 800.⏎It's an important number to
-
largest consumer of cocaine and has been leading efforts to eradicate coca production in the Andes for decades.⏎Bolivia is the world's first biggest producer of cocaine
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 don't *f * *f * *f * *f * *f * *f * *f * *f * *f * *f * *f * *f * *f * *f * *f *
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
and feeling?⏎Piper urges us to think for the glory of God. He demonstrates from Scripture that glorifying God with our minds and hearts is not either-or, but both-and. Thinking carefully about God fuels passion and affections for God.
-
it was written before the political philosophy known as Zionism was invented.⏎It is perhaps a pity that Lloyd George did not live long enough to see the fruition of the policy in the middle east that he and B
-
value of their hips, they include Rosa Rugosa (Japanese Rose), Sweetbriar (a.k.a. Shakespeare's Eglantine), and Rosa Blanda (Prairie Rose).⏎Probably originated by Dutch growers in 17
-
phalia and Bavaria continue to reverberate. They have once again raised our awareness of the fact that climate change is not just a vision of the future, but has already long since begun – that we can no longer afford to continue our
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
- L21:478 suppresses
- L20:1764 suppresses
- L19:2294 suppresses
- L19:2080 suppresses ampersand joining two items
- L16:788 excites
- L20:1194 suppresses
- L21:870 excites
- L19:1667 excites
- L20:54 weakly suppresses
- L21:1380 weakly suppresses
- L19:214 weakly suppresses
- L21:583 weakly suppresses
- L4:57 weakly suppresses
- L12:3057 weakly excites
- L13:2192 weakly excites
- L17:473 weakly excites
- L18:1167 weakly excites
- L11:2733 weakly suppresses
- L6:2867 weakly excites
- L7:92 weakly suppresses
- L17:2269 weakly excites
- L10:39 weakly excites
- L21:15 weakly excites
- L10:797 weakly suppresses
Feeds later layers
- L27:2009 strongly excites
- L25:637 excites
- L23:89 weakly excites
- L25:801 weakly excites
- L27:1647 weakly suppresses
- L27:1280 weakly excites
- L23:292 weakly excites
- L23:1549 weakly suppresses
- L23:1127 weakly excites
- L27:1357 weakly suppresses
- L26:3045 weakly suppresses
- L26:1592 weakly suppresses
- L23:2480 weakly excites
- L23:1543 weakly suppresses 'up' in 'up to' limits
- L25:2882 weakly suppresses
- L27:776 weakly excites
- L25:2975 weakly suppresses
- L23:1665 weakly suppresses
- L25:1969 weakly excites
- L27:2824 weakly suppresses
- L24:278 weakly suppresses
- L27:884 weakly excites
- L26:3071 weakly suppresses
- L27:951 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
- yet
- yet
- Yet
- Yet
- 且
- ainda
- yeti
- aún
- 而又
- 而且
Suppresses
- anny
- insic
- aps
- utter
- arest
- [:-
- uty
- NS
- azo
- ury