'born' after a person's 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. 3 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 word "born" when it follows a person's name in a biography, as in "Mohammad Hatta was born in...". Turning it up made the model write in a looser, rambling way that did not clearly relate to the prompt.
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.
-
leaving people in the dark about one of the most beloved leaders India has ever produced.⏎Life history of Subhash Chandra Bose⏎Subhash Chandra Bose was born on 23 January, 1897 in Cuttack (
-
of Trebizond⏎- (1396-ca. 1472; Latin name, Trapezuntius)Humanist and philosopher, born on the island of Crete. He settled in Venice in
-
or just ‘Diksitar’. He was the son of Rāmasvāmi Diksitar and Subbalakṣmi Ammāl, born as a result of the grace of the Divine Mother Bālāmbik
-
movement leading to its independence in 1945. He was a champion of non-alignment and of socialism grounded in Islam.⏎Mohammad Hatta was born in Bukittinggi, West Sumatra, Indonesia, on August 12
-
suffering engine failure over the Gulf of Xeros. For Brown he was shot down over Germany. But both remained determined to continue flying. Arthur Whitten Brown, born 1886, was, after WW1 determined to develop his navig
-
before him. The instruments were magnetic telegraphs, and their inventor and operator was Dr. Israel T. Nicklin.”⏎Israel Thomas Nicklin was born in Loudon County, Virginia, August 12, 179
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: A 200 mW / 1000 MΩ / 5000 MVA rms voltage source, 10 A / 10 kW / 10000 kW / 50
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
coming from Earth. It’s what our entire home planet looks like from a distance of 6.4 billion kilometres.⏎It inspired the famous planetary scientist Carl Sagan to wrote his 1994 book, Pale Blue Dot: A Vision
-
themselves in Asia Minor along with other groups of Turks.⏎Following the defeat of the Seljuks by the Mongols in 1293, Othman emerged as the leader of local Turks in the fight against the tottering Byzantine
-
14 and 24 to take responsibility for their driving behavior. Skill practices and on-the-spot defensive driving techniques help change bravado to confidence.⏎DDC-Alive at 25 teaches young adults that:⏎* People in their age
-
part of Arabia; and all of North Africa through Algeria.⏎The Early Empire, 1300-1481⏎The dynasty that Othman (1258-1326) founded was called Osmanli
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
- L1:1674 excites
- L2:111 suppresses
- L0:1301 weakly suppresses
- L0:1224 weakly suppresses
- L0:178 weakly suppresses
- L1:2795 weakly excites
- L1:388 weakly suppresses
- L0:1 weakly suppresses
- L2:75 weakly suppresses
- L0:1757 weakly excites
- L0:2919 weakly suppresses
- L1:2036 weakly suppresses
- L2:2018 weakly suppresses
- L0:383 weakly suppresses
- L2:1258 weakly excites
- L2:238 weakly excites
- L2:1435 weakly excites
- L2:41 weakly suppresses
- L1:861 weakly suppresses
- L2:2557 weakly excites
- L1:2872 weakly suppresses
- L1:1855 weakly suppresses
- L0:934 weakly excites
- L0:1983 weakly suppresses
Feeds later layers
- L4:2230 excites
- L5:2975 excites
- L6:2198 weakly excites
- L26:1701 weakly excites
- L5:59 weakly excites
- L19:1890 weakly excites verb right after 'will' or 'we'll'
- L4:151 weakly suppresses
- L16:531 weakly suppresses
- L9:59 weakly suppresses "a" meaning per (a day, a year)
- L11:1892 weakly suppresses
- L9:731 weakly excites
- L11:786 weakly suppresses
- L14:2998 weakly excites
- L27:1314 weakly excites
- L19:2692 weakly suppresses
- L19:2638 weakly excites
- L7:463 weakly suppresses
- L17:1847 weakly excites
- L5:763 weakly suppresses
- L25:283 weakly suppresses
- L27:875 weakly excites
- L7:1326 weakly excites
- L9:2360 weakly suppresses
- L12:83 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
- 崽
- rts
- $("<
- uries
- opening
- opening
- (right
- 您的孩子
- לת
- анс
Suppresses
- hol
- ritic
- estic
- doi
- explo
- study
- 研究
- |⏎
- طبق
- fulfilled