'of' after 'died' (cause of death)
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. 1 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 'of' after 'died', in phrases like 'died of plague' that name what someone died from. Turning it up did not make the model talk about illness, so the effect seems limited to that phrase.
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 civil war then erupted. In the midst of this war, King John died of dysentery on October 18, 1216. His
-
, Lyon and his apprentice left town, taking a ship to Delaware. Two days after their arrival, the apprentice died of yellow fever. Reading the newspaper while he was away, Lyon was interested in accounts
-
She carried test tubes of radioactive materials in her pocket and kept them in her desk drawer. She often wrote about the pretty blue light it gave off. She died of aplastic anaemia – a direct result of her work.⏎William Bullock invented
-
quality of life of children. Early detection and treatment would greatly increase the survival chances of children with birth defects, diseases and other health deficiencies.⏎“If a child dies of a preventable cause, it is a huge tragedy and the nation has to get
-
tough one, hitting people over 65 the hardest but also putting many young adults and children into the hospital. So far, 37 children have died of flu, the Centers for Disease Control and Prevention reports. Half of the people hospitalized
-
Reformers during the French religious wars, Mercier was obliged to take refuge in Venice in 1567 and, after returning to France, he died of the plague. One of his best-known works was the Libellus de abbrev
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: I can't be sure to get the right. But I can't be sure to get the right. But I can't be sure to get the right. But I can't be sure to get the right. But I can't be
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
D level is above 30-40 ng/ml.⏎Pneumonia is yet another bacterial infection that vitamin D can help prevent. For example, during the 1918-19 epidemic influenza in the United States,
-
more than one edge. So when I draw this graph on my 2D screen, it can happen that edges
-
This was a pencilled entry only, the vehicle became replacement "N 3".⏎Some N wagons were passed onto line contractors as replacement vehicles. This occurred when contractors wagons were running on government lines and were destroyed in derailments or collisions
-
electronic devices are becoming more powerful while shedding or maintaining size. These smaller gadgets require significant amounts of energy, which current batteries don't exactly efficiently provide yet, especially after extended use. Scientists from the University of Central Florida (U
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
- L0:1224 suppresses
- L0:1301 suppresses
- L0:178 suppresses
- L0:1468 weakly suppresses
- L0:253 weakly suppresses
- L0:2998 weakly suppresses
- L0:1632 weakly excites
- L0:1492 weakly excites
- L0:1857 weakly excites
- L0:288 weakly excites
- L0:977 weakly suppresses "like" after "you'd" or "something"
- L0:2917 weakly excites
- L0:139 weakly excites
- L0:149 weakly excites
- L0:978 weakly excites
- L0:684 weakly suppresses
- L0:2543 weakly suppresses
- L0:2782 weakly suppresses
- L0:2727 weakly suppresses
- L0:1599 weakly suppresses
- L0:1121 weakly excites
- L0:2169 weakly excites
- L0:408 weakly excites
- L0:2923 weakly suppresses
Feeds later layers
- L19:2031 weakly excites line breaks in web footers and metadata
- L21:276 weakly excites
- L3:1176 weakly suppresses
- L7:488 weakly excites
- L18:2025 weakly suppresses
- L22:246 weakly suppresses
- L12:1580 weakly excites
- L18:1816 weakly excites
- L23:2420 weakly suppresses
- L9:1152 weakly excites
- L21:659 weakly excites end of document boundary
- L8:2265 weakly suppresses
- L13:405 weakly suppresses
- L9:1438 weakly excites
- L10:266 weakly suppresses
- L16:909 weakly excites
- L18:1938 weakly excites
- L4:805 weakly excites
- L8:2757 weakly suppresses
- L10:1234 weakly suppresses
- L12:2769 weakly excites
- L11:2988 weakly excites
- L5:2959 weakly suppresses
- L20:70 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
- abal
- arton
- aje
- 除此
- hift
- 縮
- estre
- ages
- age
- ứ
Suppresses
- inas
- DCF
- SOR
- cocks
- 日在
- pray
- liest
- ưới
- tabpanel
- iners