opening parenthesis after a title or name
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. 12 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 an opening bracket that follows a title, a name or a study label, like the '(' in 'Friends (NBC)' or a citation. Turning it up made the writing wander without a clear topic.
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.
-
no sex, or no television. Afterwards, the students answered questions about their sexual attitudes and beliefs.⏎Taylor used video clips from six television programs: "Friends" (NBC), "Just Shoot Me" (NBC), "Ally McBeal
-
the philosophy of Plato and the strong defense of Aristotle expressed in several works, especially his Comparatio Platonis et Aristotelis / Comparison of Plato and Aristotle (1458). This book accused the humanist circle patronized by the
-
after first sex, the great majority of teens don’t," says Meier.⏎Meier’s study, “Adolescent Sexual Activity and Well- Being" (funded in part by the National Institute of Child Health and Development), examined psychological
-
Flood, J., Heath, S.B., & Lapp, D. (Eds.), Handbook of research on teaching literacy through the communicative and visual arts (Vol 2, pp. 4
-
in stigma attached to the term intellectual disability in comparison with the previous term, mental retardation. An online questionnaire, based on the Multidimensional Attitudes Scale (Findler, Vilchinsky, & Werner, 2007),
-
of Science degree in Computer Science. This program is growing to meet the national demand for CS majors, and has two major concentrations: A. programming and software engineering (we teach Python, assembly, C, C++, Java, Matlab, VHDL
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: it's a list of common questions and a list of resources for the latest updates on the Internet is a list of references for the "Sino-English" "Baidu Baike" and "Baidu Baike" is the website
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
00 wounded.⏎- Richter, G (1870). Manual of Coorg. Stolz. p. 337. Retrieved 14 August 2014.⏎- Archer, Jeremy. "Sir
-
2-04-15. http://web.mit.edu/newsoffice/1992/citation-0415.html. Retrieved 200<|endoftext|>Our herd of awesome goats clears land and
-
grams (helpful to know if you're counting carbs).⏎Protein - For Protein, one serving of this food has a total of 1 gram.⏎Minerals - In
-
serving size available.⏎- These nutrition facts came directly from the USDA or manufacturer/restaurant.⏎- If you're using a calorie counter, remember that Fat, Carbs, and Protein calories are just close estimations based on the Atwater factors:⏎
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
- L10:271 weakly excites
- L4:701 weakly suppresses
- L6:1389 weakly excites
- L9:20 weakly suppresses
- L7:2684 weakly suppresses
- L8:305 weakly excites
- L9:188 weakly excites 'of' in 'National Institute of' names
- L11:1019 weakly suppresses
- L11:785 weakly suppresses
- L9:2406 weakly suppresses
- L6:350 weakly suppresses
- L11:216 weakly excites
- L2:1519 weakly suppresses
- L3:1421 weakly suppresses
- L10:1225 weakly excites
- L10:1039 weakly suppresses
- L9:341 weakly suppresses
- L0:1962 weakly suppresses
- L7:343 weakly excites
- L6:990 weakly excites
- L9:1414 weakly suppresses opening parenthesis after copyright or year
- L4:1428 weakly excites
- L7:1315 weakly excites
- L8:1469 weakly suppresses
Feeds later layers
- L13:1324 weakly excites
- L15:1024 weakly suppresses
- L15:2407 weakly excites
- L13:303 weakly suppresses
- L13:1148 weakly excites
- L18:2253 weakly excites
- L16:1039 weakly suppresses
- L15:498 weakly suppresses
- L17:3005 weakly suppresses
- L16:1131 weakly excites
- L27:1954 weakly suppresses
- L14:1963 weakly excites
- L15:2922 weakly excites
- L21:2041 weakly suppresses
- L18:2912 weakly excites
- L14:174 weakly suppresses
- L17:1887 weakly suppresses
- L21:560 weakly excites
- L14:1843 weakly excites
- L16:325 weakly suppresses
- L18:2974 weakly excites
- L18:1492 weakly excites
- L15:2251 weakly excites
- L13:313 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
- acting
- الخام
- bearing
- supremacy
- Each
- _)⏎
- ._⏎
- **⏎
- or
- evity
Suppresses
- posables
- asions
- ustom
- cket
- auses
- 远程
- urret
- conforms
- zeich
- życ