commas before a continuation clause
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. 13 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 small word or comma that sits just before a clause finishes or a new part of a sentence begins, such as 'and', 'but' or 'which', in long sentences. Turning it up or down gave loose, wandering text with no clear theme.
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.
-
biologist at the University of Arizona in Tucson.⏎"They had no idea what it ate. But perhaps they knew that jumping spiders were cat-like in their movements, and [they] decided to name the spider after the agile panther Bagheera in
-
who We carried (in the Ark) with Noah, and of the posterity of Abraham and Israel of those whom We guided and chose. Whenever the Signs of (Allah) Most Gracious were rehearsed to them, they would fall down
-
over future developments in NT may halt this new field are also valid. Some community nervousness is certainly justified since “these technologies offer the potential for great benefit, [but] it is not benefit unalloyed.”12⏎However,
-
to great violence, with swelling, soreness to the touch, and sometimes redness of the skin…. This is a very formidable and extremely painful disease, [and] generally continues for many weeks…. The consequences of this disorder are often painfully
-
amyotrophic lateral sclerosis (ALS) and Huntington’s disease, but don’t always activate the right proteins at the right time, a recent study shows.⏎“[The cells] do not always realize there is a problem, or know how to
-
[and] having a forest products industry that begins to move toward oaks. If these come true, that's going to change the nature of the industry and [it] will [have to] adapt. [Awareness of climate change]
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 need to be honest about the fact that the information is too accurate for the future of the world. The information is accurate for the future of the world. The information is accurate for the future of the world. Information accurate for the world
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
z, 2011.⏎"Purchasing Power Parity Exchange Rates for the Global Poor,"⏎American Economic Journal: Applied Economics,⏎American Economic Association, vol. 3(2), pages 137-66, April
-
manner suggesting that they play a role in setting up or maintaining the 180 degrees rotational symmetry of the two flagella.⏎Rights and Permissions⏎Citation: J Cell Biol. 1983 Sep;97(3):
-
discussion of Mencius' views, the development of his ideas in the Sung and Ming and by contemporary Neo-Confucians is also considered|⏎|Keywords||No keywords specified (fix it)|⏎|Through your library||Configure|⏎Similar
-
important points of With Clause in oracle. In following section i would like to explain about the syntax as well as multiple examples of with clause.⏎With SQL_Query_Name As⏎Select * from New_SQL_Query_name;⏎Execution of With Clause :⏎In this
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
- L12:2933 suppresses
- L9:2279 weakly excites
- L9:1443 weakly suppresses
- L11:2191 weakly excites unclear: brackets and parentheses in text
- L8:4 weakly excites
- L9:2328 weakly excites
- L4:1947 weakly suppresses
- L7:174 weakly suppresses
- L7:933 weakly excites
- L8:7 weakly excites
- L12:1172 weakly suppresses
- L8:10 weakly excites
- L6:2894 weakly excites
- L8:1332 weakly suppresses
- L6:801 weakly excites
- L6:2454 weakly suppresses
- L7:2485 weakly excites
- L6:2783 weakly excites
- L9:2143 weakly excites
- L3:2734 weakly suppresses
- L7:1575 weakly suppresses
- L3:610 weakly excites
- L12:804 weakly excites
- L12:1826 weakly suppresses
Feeds later layers
- L14:2191 excites
- L14:774 suppresses
- L16:1447 excites
- L18:2120 excites
- L18:1247 weakly suppresses
- L22:986 weakly suppresses
- L22:637 weakly excites
- L19:2371 weakly suppresses
- L19:1678 weakly excites
- L16:2805 weakly suppresses
- L16:52 weakly excites
- L20:1096 weakly suppresses
- L16:2350 weakly suppresses
- L26:1824 weakly suppresses
- L20:1733 weakly suppresses
- L14:2664 weakly excites
- L14:1810 weakly excites
- L15:3030 weakly suppresses
- L20:1281 weakly suppresses
- L22:2921 weakly suppresses
- L23:2242 weakly suppresses
- L19:542 weakly suppresses
- L26:356 weakly suppresses
- L16:892 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
- ot
- 书
- �
- inge
- LAB
- potentially
- .language
- jąc
- Zend
- .bind
Suppresses
- eel
- ategorical
- igsaw
- (super
- 舄
- 世界第一
- trys
- rious
- hec
- ivering