'and' or comma between paired items
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. 13 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
Fires on the word 'and' or a comma that sits between two parallel names or items, such as 'officers and non-officers', 'David and Michael' or 'modelspace and paperspace'. It tends to go quiet on the space inside a number, such as 'an hour and 15 minutes'. Turning it up produced repetitive text, so the steering does not clearly confirm it.
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.
-
the Tata Employees Union (TEU) formed in 1935. There are two ranks of the workforce at Tata Steel, i.⏎e., officers and non-officers (workers who work in mines and plants and staff members). TW
-
d.1759 Madison County, Virginia⏎Facts and Events⏎About Mathias Cristler⏎Matthais and his two younger sons, David and Michael removed to the land grant located in the center of the present city of Madison
-
under power relays, directional or reverse current relays differential relays, and distance relays.⏎According to timing characteristics the relays can be divided into instantaneous, definite time lag, inverse time lag and inverse definite minimum time (IDMT)
-
it in the START menu on the bottom left corner of your computer screen.Ad⏎2Go to modelspace. There are two views in AutoCAD: modelspace and paperspace. Your drawing should always be made in modelspace, and dimensions that
-
long list should show up, displaying various toolbars containing different commands. The most popular toolbars that are used for AutoCAD 2D drawings are the DRAW, MODIFY, and OBJECT PROPERTIES toolbars. Select these toolbars<|endoftext|>February
-
RNA-protein complexes called small nuclear ribonucleoprotein⏎particles (snRNPs). The RNAs found in snRNPs are identified as U1, U2, U4, U5 and U6 (snRNAs).
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: the power of a community, but I have no means of doing it, but I can't do this, but this means is a community, but I have no means of doing this, but this is a community, but this is a
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
adults get at least 2 1/2 hours of moderate-intensity cardio exercise each week. Alternately, you can perform an hour and 15 minutes of vigorous cardio exercise. By regularly performing cardio exercise, you can
-
quartz crystal, and measured 5 inches high, 7 inches long, and 5 inches wide, weighing 11 pounds, 7 ounces. It closely corresponds in size to a small human cranium, with
-
health care provider if⏎Call 911 or go to an emergency room if:⏎- You think vomiting is from poisoning⏎- You notice blood or dark, coffee-colored material in the vomit⏎Call a health care provider right away or
-
Holiday Crowd At The Zoo⏎- The Child Is Father Of The Man (Es⏎Past Perfect Tense Interrogative⏎Had + subject + verb 3rd from + object.⏎Past Perfect Interrogative T
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
- L12:51 excites
- L10:838 excites comma in a list of paired nouns
- L6:427 suppresses
- L5:1469 suppresses
- L12:2808 suppresses
- L12:1505 suppresses
- L9:92 suppresses "and" or "or" joining two items
- L7:1223 suppresses "and" joining two decades or times
- L12:1786 excites
- L7:519 suppresses
- L8:1238 suppresses
- L6:1052 suppresses
- L6:1868 excites
- L6:1706 suppresses
- L8:70 excites 'and' between stacked superlatives
- L12:690 excites
- L8:1046 excites
- L12:1336 suppresses
- L1:1946 suppresses
- L8:2014 excites
- L9:1449 suppresses
- L10:1053 suppresses
- L8:131 weakly suppresses
- L7:498 weakly suppresses
Feeds later layers
- L14:976 suppresses
- L14:672 excites
- L16:2094 suppresses
- L15:2737 suppresses
- L22:1403 weakly suppresses
- L17:46 weakly excites
- L16:2048 weakly excites
- L14:745 weakly excites
- L14:3060 weakly suppresses
- L16:229 weakly excites
- L22:1029 weakly suppresses
- L15:1459 weakly suppresses
- L15:752 weakly suppresses
- L15:2598 weakly excites
- L15:670 weakly excites
- L17:1604 weakly suppresses
- L16:1179 weakly excites
- L14:1426 weakly suppresses
- L15:2201 weakly suppresses
- L16:316 weakly excites
- L16:300 weakly excites commas inside lists of nouns
- L19:2735 weakly suppresses
- L16:2521 weakly excites
- L16:485 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
- m
- m
- ра
- M
- M
- oga
- ht
- hm
- ствовать
- hb
Suppresses
- condem
- rooting
- setMessage
- anness
- seksi
- disciplined
- setName
- atism
- elib
- decency