comma after a sentence subject
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. 21 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 a comma that comes straight after the subject of a sentence, such as 'Nicolas Sarkozy,' or 'Rural girls,', and before the verb comes. Turning it up gave text that was not clearly related to this.
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.
-
posited answers to this question, whether directly - in the case of Plato and Plotinius - or through story - in the case of Homer.⏎Plotinus, having been a proponent of Platonism, naturally concurs with Plato,
-
François Mitterand who, in 1954, defiantly pronounced that ‘L’Algérie, c’est la France’. Nicolas Sarkozy, on the other hand, dramatically distanced himself both from Chirac and from
-
Manhattan Project, decided that they would declare an American monopoly over the atom.⏎Logically enough, support for this idea was not universally shared. Great Britain and France, two nations with long histories of nuclear science and technology were determined to make progress in
-
four years.⏎The disparity is partly due to the immense impact of rural students (by comparison, the urban poor are in school for over six years). Rural girls, however, get only three years of education, on average.⏎While there are a
-
influence of the ducal court and supporters of the powerful bishops of Toul, Metz and Verdun. Numerous artists came from Flanders, Germany and, sometimes, Italy to work for the Dukes of Lorraine and their friends.
-
decade and perhaps longer, the growth path and survival rate of new businesses means they are generating fewer new jobs.⏎The businesses that started in 2009, for example, are on course to create 1 million fewer jobs in the next
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,,e,.,,e,.,,e,.,,e,.,,e,.,,e,.,,e,.,,e,.,,e,.,,e,.,,e,.,,e
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
may be happy with a confidence level of 95 percent (giving a margin error of 0.05), or you may require a tighter accuracy of a 98 percent confidence level (a margin of error
-
(both images from Fri Aug 9 2019)⏎“Similar particles have been detected near Saturn, suggesting⏎that the process is at work there as well,” says Liu. “Indeed,” adds⏎Angelopoulos, “this type
-
upon it) wreaks may not be the only way that complexity grew.⏎Protein flaws called “dehydrons” may have made proteins less stable in water and made them more likely to end up working together, building up complex function.⏎Michael
-
which can significantly increase a person’s risk of developing diabetes, he says.⏎However, there is evidence that if you lose weight, you have fewer dying fat cells and significantly fewer fat-engorged macrophages, Green
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
- L20:1441 strongly suppresses
- L20:82 excites
- L19:934 excites
- L18:174 excites
- L20:663 suppresses
- L20:1358 suppresses
- L18:2376 excites
- L20:2730 excites
- L20:2674 suppresses
- L18:1791 suppresses
- L19:1018 excites
- L17:1711 excites
- L18:2593 excites
- L17:723 suppresses
- L19:949 excites
- L20:1493 excites
- L19:2999 excites
- L19:2025 excites comma before 'however'
- L19:682 suppresses
- L20:2788 excites
- L17:1845 excites
- L19:845 suppresses
- L19:650 suppresses
- L19:767 excites
Feeds later layers
- L26:236 strongly suppresses
- L27:86 strongly excites
- L26:22 strongly excites the 'res' fragment inside words
- L27:358 strongly suppresses
- L22:368 strongly suppresses
- L27:130 strongly excites
- L25:328 strongly suppresses
- L27:191 strongly suppresses
- L24:216 strongly suppresses
- L25:438 strongly suppresses
- L22:158 strongly excites
- L25:1221 strongly suppresses
- L27:286 strongly suppresses
- L26:80 strongly excites
- L26:345 strongly suppresses
- L24:572 suppresses
- L22:940 suppresses
- L24:304 suppresses
- L27:77 excites
- L26:307 excites
- L24:32 suppresses
- L26:844 excites
- L27:139 excites
- L24:41 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
- lest
- ")!=
- ourselves
- "),"
- ")==
- лиш
- "]=="
- ");//
- ']]['
- /Desktop
Suppresses
- ,
- ,
- ,',
- ,C
- ,...⏎⏎
- ,F
- ,...⏎
- ،
- ,M
- ,S