unclear: 'High' in school names (negative)
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. 5 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them 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 is hard to pin down. Its strongest dips come from the word 'High' in school names like 'Bemidji High School', while its positive peaks are scattered across unrelated words such as 'Urban' in 'Housing and Urban Development', 'canton' and 'native'. Turning it up mostly made the text wander off topic, so no clear everyday theme shows up.
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.
-
.<|endoftext|>Did you know that at superspeedways like Daytona, two or more cars running nose-to-tail can run faster than one?⏎On the racetrack, both drag and downforce are affected by the air flow off nearby cars.
-
: Mark D. Shroder⏎- Associate Editor: Michelle P. Matuga⏎Measuring Housing Affordability⏎U.S. Department of Housing and Urban Development⏎The views expressed in this article are those of the author and do not
-
and patriotism? This becomes a particularly good question, if not a surprising one, because, historically speaking, ideas of nationalism and patriotism have never emerged from military cantonments, the soldier’s barracks or from a general’s writings. If anything,
-
as you will often notice, there are some basic differences and similarities in each of these cases and it comes down to a number of language writing and speaking rules.⏎Native English Hacks and techniques for foreign students⏎A desire to start writing English like
-
2009 (for more, read below); and,Laikipia region of Kenya (January, 2009), a German canton of Switzerland (March, 2009), and the Piemonte
-
of combining content and language learning (CLIL) comes from countries such as the USA and Australia, where many immigrant people are forced to learn English as the second language, so that, they will be accepted in the society. In addition, it
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 fact that the city has a lot of power to the point of the city's highest education system, and the college has a lot of power to the point of the city's highest education system. The city's highest education system has a
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
's population as you watch the video program.⏎(This activity is based in part on a lesson by Artis West of North Cobb High School in Kennesaw, Georgia. That lesson plan is included in Featured Lesson
-
The World As It Could Be Tree Dedication Ceremony, students, teachers, and other collaborators on the project reflect on the Universal Declaration of Human Rights. Balboa High School Tree Ceremony from BAYCAT.<|endoftext|>Several images have been returned to Earth
-
Nancy Riestenberg, School Climate Specialist at the Department of Education.⏎Riestenberg joined members of the task force to speak with students from Bemidji High School and students from Schoolcraft to discuss their concerns about bullying and ways in which
-
ujillo grew up in Corcoran, CA and was one of seven children in a family who earned a living as farm workers. While at Corcoran High School,
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
- L2:1564 excites
- L4:2822 excites
- L3:1950 excites
- L3:1244 excites
- L3:1216 weakly suppresses
- L4:182 weakly suppresses
- L4:23 weakly excites
- L4:215 weakly suppresses
- L3:1437 weakly excites
- L4:25 weakly excites
- L4:19 weakly suppresses
- L4:2927 weakly suppresses
- L4:21 weakly excites
- L3:684 weakly suppresses
- L3:2434 weakly excites
- L3:917 weakly suppresses
- L3:1831 weakly excites
- L1:2782 weakly excites
- L2:627 weakly excites
- L1:1111 weakly suppresses
- L4:1818 weakly excites
- L4:989 weakly suppresses
- L4:2123 weakly excites
- L3:1889 weakly excites
Feeds later layers
- L6:719 suppresses
- L7:724 excites
- L7:272 suppresses
- L8:265 suppresses
- L10:2102 suppresses
- L6:307 suppresses
- L10:404 excites
- L11:598 excites
- L7:845 suppresses
- L10:2438 suppresses
- L9:181 suppresses
- L7:267 weakly suppresses
- L7:1054 weakly excites
- L15:1012 weakly suppresses
- L6:1639 weakly excites
- L15:715 weakly suppresses
- L8:2137 weakly suppresses
- L7:1196 weakly suppresses
- L9:1341 weakly suppresses
- L24:671 weakly suppresses
- L10:892 weakly excites
- L12:1184 weakly suppresses
- L13:2360 weakly suppresses
- L16:2835 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
- icing
- onet
- 也有一些
- тор
- ront
- jednoc
- 一个是
- rende
- enge
- Di
Suppresses
- Var
- arlo
- 产业集聚
- atters
- Var
- -dismiss
- ndata
- houses
- Pvt
- embre