inside web addresses
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 mixed: they peak on different tokens. 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 parts of website addresses, such as the 'bc' in bbc.co.uk, the '.com' after a company name, or the '.edu' in a university address. Turning it up did not make the model write links, so the effect in the output is unclear.
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.
-
your quality of life.⏎Think Positive. Nothing is lost when you maintain hope.⏎BBC news . https://www.bbc.co.uk/news/health-49999404⏎‘
-
you to use them however you feel will work best for your child and your own personal circumstances.⏎Check out the full range of plans here: https://www.edplace.com/blog/tag/learning_plans.⏎What are the EdPlace learning plans?⏎
-
research fellow at CNRS http://www.cnrs.fr.⏎Actoz is a database for social history. It is a Filemaker-based http://www.filemaker.com/ application designed to develop a database for historical analysis. It provides an ingenious and
-
Unfortunately, most of the sodium in your diet comes from processed and prepared foods, so follow these tips to shake your salt habit:⏎*http://www.washingtonpost.com/wp-dyn/content/article/2010/04/1
-
and Issues in Education. It will also be a welcome addition to individual scholars' bookshelves.⏎Meet the author! http://www.education.miami.edu/ep/vita<|endoftext|>Society is usually considered morally upright and successful when all
-
information⏎Difficulty seeing or processing/understanding what is seen⏎For additional information please contact the Counselor/Coordinator of Student Accommodations at 217/234-5259 or email Jordan Hicks at email@example
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 have been reading some books and I have been thinking about the books and I have been reading. I have been thinking about some books and I have been reading. I have been reading, and I have been thinking about some books. I
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
is that the wick is not the right size for the candle.⏎The scent released from a candle when it is not lit.⏎The scent released from a candle when it is burning.⏎Ahhhhhhhh the glorious melt pool! A full
-
. We offer recommendations for improving alignment to help students more successfully participate in math instruction across settings. Math Curriculum Crosswalk: Grade 1 Math Curriculum Crosswalk: Grade
-
a narrow waterway connecting two larger bodies of water⏎a state made up of an independent city and the nearby countryside⏎a government by a few⏎a government by the people⏎a government by one ruler⏎a group of people called into court
-
13⏎Updated: March 13, 2013⏎Updated: October 02, 2013⏎At the outset, it
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
- L12:605 weakly excites
- L11:784 weakly suppresses
- L5:50 weakly suppresses
- L7:2331 weakly suppresses
- L8:942 weakly excites
- L8:169 weakly excites
- L7:1660 weakly excites
- L11:1886 weakly suppresses
- L12:3016 weakly suppresses
- L3:115 weakly excites
- L6:514 weakly suppresses
- L12:447 weakly suppresses
- L8:297 weakly excites
- L2:2870 weakly excites
- L10:36 weakly excites
- L10:73 weakly suppresses
- L3:294 weakly suppresses
- L7:1043 weakly excites
- L3:421 weakly excites
- L12:1705 weakly excites
- L7:49 weakly excites
- L12:2812 weakly excites
- L7:616 weakly suppresses
- L12:2007 weakly suppresses
Feeds later layers
- L14:1077 excites
- L14:1241 excites
- L14:455 suppresses
- L18:1633 weakly suppresses
- L16:1047 weakly suppresses
- L14:501 weakly suppresses
- L14:2351 weakly excites
- L15:1533 weakly excites
- L15:325 weakly excites
- L24:2940 weakly excites
- L20:807 weakly excites
- L17:1144 weakly suppresses
- L14:341 weakly excites
- L15:1466 weakly excites
- L15:1504 weakly excites
- L14:393 weakly suppresses
- L14:1138 weakly excites
- L18:2035 weakly suppresses
- L14:1081 weakly suppresses
- L14:3063 weakly suppresses
- L16:790 weakly suppresses
- L18:2094 weakly suppresses
- L14:1961 weakly excites
- L15:2382 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
- ie
- /gl
- hiếu
- реб
- eways
- ないこと
- er
- rema
- 射
- emm
Suppresses
- ��索
- 矍
- ighted
- 籍
- *@
- blindness
- cala
- 磻
- .Italic
- 市教育