hyphen inside phone numbers
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. 14 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 lights up on the dash in the middle of a phone number, such as the one in 859-218-0720. Turning it up made the model's writing wander into unrelated topics, so the steering tells us little about it.
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.
-
's coat, sweet potato vine and lots of different begonia varieties as possible alternatives.⏎Nicole Ward Gauthier, 859-218-0720<|endoftext|>Sayings of Samoan origin⏎A careless person
-
vocational education, instructional materials and extra-curricular activities. Report violations of this policy to the Assistant Superintendent for Administrative Services, 703-228-6008, or the Assistant Superintendent for Human Resources, 70
-
part of the UPMC family.⏎Be safe anytime, anywhere.⏎To find a pediatrician or pediatric specialist, please call 412-692-7337 or search our directory.⏎A resource for our network of referring
-
a⏎safe and dark observing site: check out the locations where the⏎Southern Cross Astronomical Society⏎(Phone:305-661-1375) is holding public observing sessions. Also the⏎Miami Museum
-
Extension⏎Published: Monday, March 18, 2013⏎David H. Trinklein, 573-882-9631⏎COLUMBIA, Mo. – It
-
, you may have acid erosion. To schedule a dental checkup with Greg Ritchie at Ritchie Dental Group, call 830-693-8833 today. We are a full-service dental office with hundreds of
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: the history of the idea of the person, the mind, and the body, and the mind, the history of the idea of the person, and the body, and the mind, and the history of the person and the mind, the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
quite clear how these central pattern generators know how quickly the legs are to be moved.⏎An international team working with scientists from Harvard University and the Max Planck Institute of Neurobiology in Martinsried has now discovered individual neurons in the brain of zeb
-
in a glass of water, the light from the pencil in the water is being refracted, so it appears in a different location. The result is the illusion of a broken pencil
-
a charge of bad taste,<|endoftext|>8 edition of Principles of Biblical Interpretation found in the catalog.⏎April 1, 2
-
the Children’s Hospital Pediatric Telemedicine Program at the University of California, Davis.⏎“People say a picture is worth a thousand words,” says Marcin, professor of pediatrics. “With medicine, video conferencing brings us right to the bedside
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
- L13:287 suppresses
- L13:1251 excites
- L10:272 suppresses
- L13:700 excites
- L9:330 suppresses gap where a number was removed
- L10:2221 excites
- L13:68 excites
- L13:175 excites
- L10:404 suppresses
- L13:1540 weakly excites
- L13:1049 weakly suppresses
- L13:1181 weakly suppresses 'of' after a definition or noun
- L9:316 weakly suppresses
- L13:1280 weakly suppresses
- L13:977 weakly suppresses
- L12:2667 weakly excites
- L6:2119 weakly suppresses
- L13:838 weakly excites
- L8:307 weakly suppresses
- L11:31 weakly suppresses
- L9:29 weakly excites opening parenthesis before an acronym
- L13:2465 weakly excites
- L2:1106 weakly excites
- L1:1955 weakly suppresses
Feeds later layers
- L16:2297 excites
- L15:1438 excites
- L18:1727 weakly excites
- L17:1303 weakly excites
- L20:2117 weakly excites
- L20:1372 weakly excites
- L16:2662 weakly excites
- L16:1921 weakly excites
- L18:2856 weakly excites
- L20:1031 weakly excites
- L16:2660 weakly excites
- L21:945 weakly excites
- L17:777 weakly suppresses
- L21:1764 weakly excites
- L16:1072 weakly excites
- L27:2510 weakly excites
- L19:170 weakly suppresses
- L17:433 weakly suppresses
- L15:2615 weakly excites
- L16:121 weakly suppresses
- L17:2402 weakly suppresses
- L23:1578 weakly excites
- L15:815 weakly excites
- L17:1872 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
- missive
- asures
- iode
- ше
- (off
- ypse
- <dim
- olvers
- вин
- ыш
Suppresses
- _UNIQUE
- 찍
- /script
- 람
- atcher
- repetition
- التواصل
- /function
- enerator
- SCRIPT