the 4 in years 184x, 164x
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. 21 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 digit 4 inside years from the 1600s and 1800s, like 1847 or 1648. Turning it up did not produce a clear theme in the output.
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.
-
. A good place to start or to finish is at the beginning.⏎The first U.S. general-issue postage stamps were issued July 1, 1847: a 5¢ red brown Benjamin Franklin stamp (Scott 1)
-
Falls Convention was a meeting of supporters of women’s rights in the United States. The meeting took place on July 19-20, 1848, at Seneca Falls, New York. It helped to launch the U
-
the years 1642 to 1649, it tells the story of Hester P
-
ax. This spruce gum, it should be noted, had little to no flavor and was akin to chewing on a piece of wood.⏎In 1848, an American businessman named John Bacon Curtis recognized the massive financial potential of turning
-
20th century.⏎© Science Museum / Science & Society Picture Library⏎This wall telephone uses the chalk receiver invented by Thomas Alva Edison (1847-1931) in 1878. Edison redesigned
-
5 John Hancock was elected president of the second Continental Congress.⏎In 1819 Queen Victoria was born in Kensington, near London.⏎In 1844 Samuel Morse transmitted the
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 has been a little while, and I have to have a little more thanallyallyallyallyallyallyallyallyallyallyallyallyallyallyallyallyallyallyallyallyallyallyallyallyallyallyallyallyallyallyallyally
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
-polluting nations such as<|endoftext|>Learning disabilities are neurologically based conditions that interfere with the acquisition, storage, organization, and use of
-
and Statistical Manual of Mental Disorders, Third Edition, Revised criteria: (1) alcoholism with tolerance or dependence; (2) alcoholism with or without tolerance-dependence; and (3) alcoholism with or without tolerance-dependence or problem
-
is with format of⏎diff output, not just with⏎patch, because with odd enough file names one can cause⏎diff to generate a patch that is syntactically correct but⏎patches the wrong files. The<|endoftext|>Dante Gabriel Rosset
-
degree of autonomous operation. At the low end of autonomy, they could be deployed to handle traffic laws on their own. On the higher end, they could operate autonomously to conduct arrests of suspects who might resist arrest violently. Near the highest end
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
- L20:61 strongly suppresses
- L2:202 strongly suppresses "of" after "many" or "which"
- L16:41 strongly excites
- L16:3 suppresses
- L0:117 excites
- L1:259 suppresses
- L4:216 excites
- L2:761 suppresses
- L18:853 suppresses
- L0:141 suppresses the word 'of' in 'middle of'
- L16:90 suppresses
- L4:524 excites
- L4:1214 excites
- L16:39 excites
- L19:43 excites
- L2:1753 excites
- L4:345 excites
- L5:1352 excites
- L18:1997 suppresses
- L3:459 suppresses
- L2:598 excites
- L20:854 excites
- L20:959 excites
- L20:895 suppresses
Feeds later layers
- L22:2467 excites
- L26:1696 suppresses
- L24:772 weakly suppresses
- L24:834 weakly excites
- L25:3070 weakly suppresses
- L22:1055 weakly excites
- L23:474 weakly suppresses
- L23:2661 weakly excites
- L23:515 weakly excites
- L23:508 weakly suppresses
- L23:1356 weakly suppresses
- L27:244 weakly excites
- L23:1364 weakly suppresses
- L23:999 weakly suppresses "be" after modal verbs
- L23:1114 weakly excites
- L25:1381 weakly excites
- L24:979 weakly excites
- L23:1817 weakly excites
- L27:767 weakly suppresses
- L22:563 weakly excites
- L22:1987 weakly suppresses
- L27:1501 weakly excites
- L25:1731 weakly excites
- L22:331 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
- ive
- ly
- ic
- LY
- IVE
- enticated
- culate
- raries
- al
- भ
Suppresses
- ally
- als
- ality
- ials
- ically
- */;⏎
- ALLY
- ics
- (ball
- nal