comma after 'for example'
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. 15 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 comma right after phrases like 'For example' or 'For instance', just before the example itself starts. Turning it up or down did not produce a clear, matching change in the written text.
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.
-
reason)". Often, you can use exceptions to handle this situation.⏎Sometimes you might want to return objects of different types based on arguments to the function. For instance, did the caller ask for the data in this format or that format?⏎Sometimes you
-
answers.⏎- Be specific. When setting goals, do some background work. Research your child’s favorite schools together to find out the criteria for admission. For example, if their school of choice requires a 3.5 GPA but your child is
-
0. Then we use a symbol like a triangle, circle, line, star or a box to show if that number has that many equal groups. For example, on our 10th day of school we found 5 groups of
-
all AVA’s are appellations, but not all appellations are AVA’s. For example, a state or county name may be used as an appellation on a wine label
-
of juvenile court involvement among a community's young mental health clients is substantial. For example, a study in one city found that adolescents in contact with the community's mental health
-
use the numeral converter for the number of reasons in their lives. For instance, you may want to list up the number of pages in your book using the Roman
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: (1) the problem of the (1) the (1) the (1) ) the (1) ) the (1) the (1) the (1) the (1) the (1) the (1)
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
|• Land||96.6 sq mi (250.2 km2)|⏎|• Water||0.7 sq mi (1.7 km2)|⏎|
-
the same amount of time running at a pace of 5 miles per hour. A study in the May 1996 issue of the "Journal of the American Medical Association" also concluded that running on a treadmill is superior to using the
-
processing, and analyzing national security information from around the world. The CIA is an independent agency, overseen by the Director of National Intelligence, and reports to both the President of the United States and the United States Congress.⏎The CIA is an independent
-
a vegetarian diet causes weight loss, even in the absence of exercise or calorie counting, according to a new meta-analysis published as an online advance in the Journal of the Academy of Nutrition and Dietetics on Thursday, Jan. 22,
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
- L10:29 suppresses
- L10:77 weakly suppresses
- L9:22 weakly excites
- L7:1123 weakly suppresses
- L13:414 weakly suppresses
- L7:2705 weakly suppresses
- L8:2149 weakly excites
- L2:1211 weakly suppresses
- L8:127 weakly suppresses
- L2:2940 weakly suppresses
- L14:692 weakly excites word right before a section number
- L8:1648 weakly excites
- L12:488 weakly suppresses
- L14:348 weakly suppresses
- L1:990 weakly excites
- L1:591 weakly suppresses
- L1:1062 weakly suppresses
- L7:1346 weakly excites
- L8:851 weakly excites
- L13:1862 weakly excites
- L6:2182 weakly suppresses
- L11:1158 weakly suppresses
- L14:542 weakly excites
- L8:1433 weakly suppresses
Feeds later layers
- L19:94 suppresses
- L16:71 weakly suppresses
- L16:87 weakly suppresses
- L16:110 weakly suppresses
- L16:1304 weakly suppresses
- L17:628 weakly excites
- L24:2033 weakly excites
- L20:2407 weakly suppresses
- L18:432 weakly suppresses
- L23:100 weakly excites comma or full stop ending a clause
- L16:946 weakly suppresses the word 'late' in time phrases
- L16:2461 weakly suppresses
- L22:612 weakly excites
- L16:287 weakly suppresses
- L18:598 weakly suppresses
- L20:2122 weakly excites
- L22:689 weakly excites comma closing an intro clause
- L26:1625 weakly suppresses
- L16:193 weakly suppresses 'wide' in 'a wide range of'
- L16:306 weakly suppresses
- L17:243 weakly suppresses
- L27:422 weakly excites
- L16:105 weakly suppresses
- L17:203 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
- environment
- Environment
- matic
- PCS
- abar
- pies
- 环境下
- consider
- accept
- +z
Suppresses
- aurant
- pendicular
- 斯塔
- etadata
- Stuart
- Vere
- steller
- ceed
- świ
- owntown