surname split after first letter D
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 fires on the letter D when it starts a surname that the tokenizer has split in two, as in 'Jason Dworkin' or 'Gwynne Dyer'. Turning it up pushed the model toward unrelated, rambling text, so the steering does not show a clear effect.
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.
-
be available for study for future generations, it is possible the person who names the asteroid will grow up to study the regolith we return to Earth,” said Jason Dworkin, OSIRIS-REx project scientist at NASA’s Godd
-
after a baby’s birth, a period that has been dubbed the “magic moment.” The new study is the first to test that assumption, said author Christina Gibson-Davis. “It turns out the ‘magic moment’ lasts longer than conventional wisdom
-
found that most children born out of wedlock don’t remain so: 64 percent of children born out of wedlock see their moms get married, Gibson-Davis said. Many of those marriages don’t last, however. Nearly half of
-
subject. Hughes argues that Shakespeare's concern with Lucrece's chastity is reflective of a larger metaphysical concern with the sacredness of the soul. Carolyn D. Williams (1993) examines Lucrece's rape, as
-
heading into deep space.⏎“The mission was essentially over by<|endoftext|>Human nature and the nature of war⏎Twenty years ago, Canadian military historian and journalist Gwynne Dyer fascinated the public in 45 countries with his award-winning television series
-
going to be another productive year in physical education class!<|endoftext|>Spring 2010 ACCESS #36⏎Just Road Pricing by Lisa Schweitzer and Brian D. Taylor⏎Economists have long advocated road pricing as an efficient way to
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: what if I can actually actually actually actually actually actually actually actually actually actually actually actually actually actually actually actually actually actually actually actually actually actual? I have to check to see if this actual of actual of actual of actual of actual of actual of actual
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
قاله پژوهشی⏎1 دانشیار معماری، دانشکده معماری و شهرسازی، دانشگاه علم و صنعت ایران، تهران
-
شگر دوره تخصصی دکتری معماری منظر، دانشکده معماری و شهرسازی، دانشگاه علم و صنعت ایران، تهران
-
engrin⏎There are straws in the wind, a growing discontent with the French.Here Be Dragons⏎Trends of 'discontent'⏎View usage over:⏎'discontent' in Other Languages⏎Translation of discontent from the Collins English to Portuguese
-
"Here we show that Epo increases the motivation to exercise," said Max Gassmann, D.V.M., a researcher involved in the work from the Institute of Veterinary Physiology,
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
- L23:400 excites
- L22:1511 suppresses
- L2:1125 weakly excites
- L23:89 weakly suppresses
- L19:251 weakly excites
- L18:2364 weakly suppresses
- L20:259 weakly suppresses
- L23:86 weakly suppresses
- L18:569 weakly excites
- L4:2965 weakly excites
- L15:1893 weakly excites
- L14:1884 weakly suppresses
- L21:237 weakly suppresses
- L3:952 weakly excites
- L6:2812 weakly excites
- L22:2665 weakly suppresses physical measurements and dimensions
- L10:1673 weakly excites
- L10:1774 weakly suppresses
- L22:793 weakly suppresses
- L19:841 weakly excites
- L14:2928 weakly suppresses
- L20:2381 weakly suppresses
- L6:240 weakly suppresses
- L0:371 weakly excites
Feeds later layers
- L27:1955 excites
- L25:1285 suppresses
- L25:1817 weakly suppresses
- L25:572 weakly suppresses
- L25:1814 weakly suppresses
- L27:874 weakly suppresses
- L26:1396 weakly suppresses
- L25:1142 weakly suppresses weight loss and metabolism claims
- L27:2852 weakly suppresses
- L27:1796 weakly suppresses
- L25:649 weakly suppresses
- L25:2488 weakly excites
- L26:1040 weakly suppresses
- L27:1522 weakly suppresses
- L25:1418 weakly excites
- L27:1976 weakly suppresses
- L27:686 weakly excites
- L26:2020 weakly excites
- L26:2095 weakly suppresses
- L25:1703 weakly suppresses
- L27:555 weakly excites
- L25:2118 weakly suppresses
- L25:408 weakly excites
- L26:2014 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
- мы
- úa
- dın
- positories
- стер
- 再也不
- UV
- ASON
- erra
- 不做
Suppresses
- anden
- ector
- toInt
- ern
- )-(
- entifier
- 对象
- (type
- .allocate
- 拊