unclear: mid-word pieces of 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. 25 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. 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 middle of split words, mostly inside surnames and place names, such as 'Etna' and 'bighorn'. Its effect on the model's writing is unclear, since turning it up gave mixed, unrelated text.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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.
-
shooter field, Washington County, Oklahoma; 1945; OFR; 45-37; Fox, I. William; Thigpen, Claude H.; Ginter, Roy L.; Alden, George P
-
," or thought to mean "having profitable pledges." The O'Boyles were chieftains in Donegal, ruling west Ulster with the O'Donnells and the O'Doughertys. Boyles can also be found in
-
game species, pronghorn antelope (Antilocapra americana), mule deer (Odocoileus hemionus), and desert bighorn sheep (Ovis canadensis mexicana). Available information suggests that predator
-
than randomly arranged topics.⏎Specifications of Becoming a Reader: A Developmental Approach to Reading Instruction (3rd Edition)⏎|Author||Michael P. O'Donnell, Margo Wood|⏎|Number Of Pages||408|⏎Write
-
do about it?"⏎How about it?<|endoftext|>Published on Feb 20, 2013⏎CBS News RAW: A new eruption on Mount Etna occurred at its South-Eastern crater. This event generated tall lava fountains
-
also be located underneath it.⏎Mount Etna is one of the most active volcanoes in the world and is in an
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: Tame Tame Tame Tame Tame Tame Tame Tame Tame Tame Tame Tame Tame Tame Tame Tame Tame Tame Tame Tame Tame Tame Tame Tame
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
, the technology went largely unnoticed, and diesel engines gained a reputation for being dirty and "smelly."⏎In 2000, the situation began to change with a
-
1 People's Hospital in Xiangyang, Hubei province. China's family planning policy has slowed the rapid growth of population. Gong Bo / Xinhua⏎Questions arise over Earth��s⏎The world's population is expected to
-
to the problem, such as frequent dining out, drinking with business partners, toasting (urging one another to drink), and popular drinking games such as 'Wager' that encourage excessive drinking.⏎China's drinking problem is not helped by the
-
re-imperialisation. And this can only be done by reducing Russia’s territorial capabilities, says Professor Andrzej NOWAK in an interview with Mikołaj CZYŻ.⏎Mikołaj CZYŻ: Are the roots of Western
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
- L20:1594 excites
- L20:2049 weakly suppresses
- L1:2984 weakly suppresses
- L20:293 weakly suppresses
- L7:73 weakly excites
- L24:2738 weakly suppresses
- L14:2928 weakly excites
- L19:398 weakly suppresses
- L3:1905 weakly excites
- L5:284 weakly suppresses
- L9:740 weakly excites
- L10:1552 weakly excites
- L24:350 weakly excites comma before 'such as' examples
- L5:1093 weakly suppresses
- L16:1525 weakly suppresses
- L8:2620 weakly suppresses
- L14:392 weakly excites
- L12:1104 weakly excites
- L14:490 weakly suppresses
- L2:1869 weakly excites
- L8:2316 weakly suppresses
- L16:2758 weakly excites
- L6:2235 weakly excites
- L4:2015 weakly suppresses
Feeds later layers
- L27:686 suppresses
- L26:977 suppresses
- L26:453 weakly suppresses
- L27:330 weakly suppresses
- L26:454 weakly excites
- L27:3061 weakly excites
- L27:420 weakly excites
- L26:1353 weakly suppresses
- L26:1143 weakly excites
- L26:1401 weakly excites
- L27:2205 weakly excites
- L27:672 weakly excites
- L27:2246 weakly suppresses the letter r starting a word fragment
- L26:1836 weakly excites
- L27:930 weakly excites
- L26:2883 weakly suppresses
- L27:426 weakly suppresses
- L27:813 weakly excites
- L27:2227 weakly excites
- L27:1654 weakly excites
- L26:2918 weakly suppresses
- L27:2231 weakly excites
- L26:2182 weakly excites
- L27:2852 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
- /jav
- antium
- ColumnType
- .pix
- inventions
- athom
- paddingHorizontal
- .bpm
- ше
- ursos
Suppresses
- T
- T
- WA
- QA
- wa
- Accepted
- QA
- _T
- WA
- leck