'the' in historical names
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. 23 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 word "the" that sits inside a famous name, like "Ivan the Terrible" or "William the Conqueror". It points toward words for "her" and away from "he" or "him". When turned up, the text shifted toward female pronouns and when turned down it kept repeating "he".
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.
-
that is, agree to a Russian sphere of influence in Eastern Europe . . .⏎The second line is of an economic nature. It originated in trade contacts between Ivan the Terrible and Elizabeth I, the Queen
-
. In particular, polygamy is the practice of a man taking more than one wife, while polyandry is the practice of a woman taking more than one husband. Polygamy is a common model of marriage in some parts of the
-
also installed sticky mats on the floor<|endoftext|>Achilles was the greatest warrior to have ever lived. He was the Golden Warrior of the Bronze Age. Even Alexander the Great admired and emulated Achilles throughout his life, sleeping with a copy of the
-
which since 1981 has been strongly linking Germany, for example, first with the USSR and later with Putin’s Russia.⏎From the times of Ivan the Terrible till this day, Russia has been trying to present itself to the Western
-
, the day of St. Stephen the Martyr, everyone goes out and visits friends and relatives, weather permitting. One
-
saw an American F-86 being attacked by a Russian MiG-15 and went in pursuit.⏎Captain Semyon Alekseyevich Fedorets, commanding the 913th IAP (Istrebitel
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: he is a 55-year-old, he is a he, he is a he, he is a he... he is a he... he is a he... he is a he... he is a he... he is a he
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
.⏎When Hamlet is engaged in a discussion with Gertrude, they are sparring with words. She is talking about one thing, he another. She tells him he has offended his father very much (meaning Claudius, who is not
-
in the past. "I stop there yesterday," though ungrammatical, gets the message across.⏎"He is here." The agreement between the third person subject "he" and the third person of the verb "is" is basically redundant.
-
a controversial figure even within the communist world. People who were his friends and those who were opposed to him held differing opinions on him and had varying interpretations and explanations of his motives for participating in the revolutions against capitalism and imperialism all over the world.⏎
-
. After his great struggle, we got freedom on 15th August 1947. He played a vital role in our independence, so we called him as the father of the nation. All our Indians with great love and affection
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
- L22:1623 strongly excites
- L19:666 excites
- L17:704 excites
- L16:748 suppresses
- L16:377 excites
- L17:428 excites
- L18:438 suppresses
- L20:2754 suppresses
- L15:342 weakly suppresses
- L18:1039 weakly suppresses
- L20:2033 weakly excites
- L21:2560 weakly suppresses
- L22:520 weakly excites
- L6:1492 weakly suppresses
- L22:1981 weakly suppresses
- L14:1022 weakly suppresses
- L19:1375 weakly suppresses
- L5:2607 weakly excites
- L3:2041 weakly suppresses
- L12:1523 weakly excites
- L11:1647 weakly excites
- L1:911 weakly excites
- L17:739 weakly suppresses
- L21:2539 weakly excites
Feeds later layers
- L27:578 strongly suppresses
- L26:869 strongly suppresses word before him or he
- L24:477 suppresses
- L24:2385 suppresses
- L27:813 excites
- L27:1244 suppresses
- L26:1143 weakly excites
- L24:1120 weakly excites
- L24:1050 weakly excites
- L24:1615 weakly excites
- L25:597 weakly suppresses
- L26:1127 weakly suppresses
- L27:857 weakly suppresses
- L27:1190 weakly suppresses
- L25:310 weakly suppresses
- L27:1865 weakly suppresses
- L27:563 weakly suppresses
- L25:2043 weakly suppresses
- L26:1061 weakly excites
- L27:1026 weakly suppresses
- L26:1038 weakly suppresses
- L25:527 weakly suppresses
- L26:2320 weakly suppresses
- L26:2643 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
- her
- 她的
- 她
- 她在
- 让她
- 但她
- 她们
- 跟她
- 她说
- Her
Suppresses
- him
- 他
- he
- 他会
- 他对
- 他也
- 他又
- 他就
- his
- 所以他