the digit 1 starting years
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. 24 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 lights up on the digit 1 when it starts a year in the 1700s or 1800s, such as the 1 in 1815 or 1766. Turning it up made the model's writing wander into odd, unrelated topics like social media and the weather, so the steering result doesn't clearly reflect the 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.
-
ing boats which were managed by watermen. These boats were known as Wherries or water taxis as we would call them.⏎Steam boats were introduced in 1815 and by 1850 were carrying more than a million
-
Spanish Empire and the rise of America as a major world power with several new possessions.⏎Britain Repeals the Stamp Act⏎On March 18, 1766, British Parliament repealed the Stamp Act. Passed a year earlier,
-
catcher mounted on the back of Daniel Webster.⏎The print might have been produced in Boston, a center of bitter opposition to the Fugitive Slave Act in 1850 and 1851.⏎Please post your comment below to
-
limits a just and liberal policy, and to give that humane and considerate attention to their rights and their wants,” Jackson declared in his first inaugural address in 1829
-
and how she came up with the idea, and the sequence of events?”⏎A Dark and Stormy Night⏎The story begins, literally, in June 1816 at Villa Diodati overlooking Switzerland’s Lake Geneva. Here,
-
ham, Lt. Col. of the 26th Illinois Volunteers. Born in Covington, Kentucky, Tinkham graduated from West Point Academy in 1849, and worked as a civil engineer. He later settled in Homer
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 truee, not the C: C A: A C: C A: A C: C A: A C: C A: A C: C A: A A
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
before the destination.⏎I will depart to mine own land. --Num. x. 30.⏎Ere thou from hence depart. --Milton.⏎He which hath no stomach to this fight
-
a politician, to look good rather then actually be good. ” to thine own self be true, And it must follow, as the night the day, Thou canst not then be false to any man. Farewell; my blessing season
-
thou lettest me steal a loaf, brandy, a hen, a goose, a pig or a horse, I will give thee a big candle, because thou art my sweet little golden God.<|endoftext|>The Hogshooter field, located in
-
his first soliloquy in Act 1 Scene 2 where he recognizes that "self-slaughter" is against God's commandment (i.e. Thou shall not kill). Here, Hamlet is being over dramatic: youthful hyperbole
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
- L23:415 excites
- L20:1391 excites
- L18:967 suppresses
- L3:2783 suppresses
- L6:1927 weakly suppresses
- L22:258 weakly suppresses
- L22:714 weakly suppresses
- L8:1492 weakly suppresses
- L0:2159 weakly suppresses
- L19:2941 weakly suppresses
- L3:187 weakly excites
- L3:2540 weakly suppresses
- L19:119 weakly suppresses
- L20:1074 weakly excites
- L0:2999 weakly suppresses
- L7:92 weakly suppresses
- L16:1543 weakly suppresses
- L0:587 weakly excites
- L21:2002 weakly excites
- L11:223 weakly suppresses
- L14:709 weakly suppresses
- L2:457 weakly suppresses
- L21:2656 weakly suppresses
- L2:2645 weakly excites
Feeds later layers
- L27:785 weakly suppresses
- L25:407 weakly excites
- L26:2776 weakly suppresses
- L27:2247 weakly excites
- L25:2029 weakly excites
- L27:763 weakly excites
- L25:108 weakly suppresses
- L26:1088 weakly excites
- L27:545 weakly suppresses
- L25:1139 weakly excites
- L26:294 weakly suppresses
- L25:2596 weakly excites
- L26:267 weakly suppresses
- L27:720 weakly suppresses
- L26:3019 weakly suppresses
- L27:1356 weakly excites
- L25:457 weakly excites
- L25:2499 weakly excites
- L25:2274 weakly excites
- L25:1895 weakly excites first digit of a year like 2020
- L25:2025 weakly excites
- L25:125 weakly excites
- L26:1353 weakly suppresses
- L25:1879 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
- ronics
- asses
- ró
- 鸠
- ạm
- Sans
- cream
- lace
- ié
- فصل
Suppresses
- bserv
- e
- ele
- 地上
- eus
- ulis
- ElementType
- 新三
- ely
- esign