job and work words
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. 25 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 tends to fire on words about jobs and work, such as 'job', 'work', and 'workbooks', as well as the word 'City' in a university name. Turning it down made the model keep repeating the word 'class', which this neuron appears to hold back.
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.
-
1 (1983): 4–43.⏎Wood, Christopher S. Albrecht Altdorfer and the Origins of Landscape. Chicago: University of Chicago Press, 1993.⏎Saint George and the Dragon
-
classroom settings, of course. Inattentive mistakes can show up in the workplace, in organizing your day-to-day life – or in particularly inconvenient settings like job interviews.⏎And people can always take those mistakes to be representative of your overall competence
-
ayashi is Associate Professor in the Department of Media and Communication at City University of Hong Kong.⏎Atsushi Tago is Professor in the School of
-
them. This makes the process of study planning and career development more concrete and measurable.⏎Students can report from their interests, career goals and experiences in internships and work. This helps explore possible career paths and formulate concrete steps to achieve them.⏎SL
-
jobs available to African Americans. The plan, which targets communities with persistently high unemployment, includes three main components: creation of public sector jobs, job training with job-placement programs, and wage subsidies for employers. Although the plan is constructed with African
-
in all five boroughs of NYC, Yonkers, Westchester, Nassau and Suffolk.⏎CITE PD: CITE offers CTLE-approved in-school or online professional development tailored to your school’s needs and your
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: class class class, class class class, class class class, class class class. Class class class, class class class, class class class, class class class, class class class. Class class class, class class class, class class class,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
half a century.⏎On June 15, 1977, former slave Henry Ossian Flipper graduated from West Point with the ranking of second lieutenant in the all-African-American 10th Cavalry.⏎While he
-
coat belonged to Charles Johnston Tinkham, Lt. Col. of the 26th Illinois Volunteers. Born in Covington, Kentucky, Tinkham graduated from West Point Academy in 1849, and worked as a civil
-
: Institutional⏎American Society of Civil Engineers Award of Merit 2015⏎Civic Trust Award 2014⏎Civic Trust Awards Special Award for Sustainability 2014⏎ENR Best Global Education/Research
-
commandant" or "adjutant," and to be composed of two companies, each consisting of a captain, one lieutenant, two ensigns, one first sergeant,
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
- L7:92 suppresses
- L20:2742 suppresses
- L9:207 weakly excites
- L3:130 weakly suppresses
- L16:1220 weakly suppresses
- L4:3038 weakly suppresses
- L23:798 weakly excites
- L1:25 weakly suppresses
- L6:1673 weakly excites
- L11:39 weakly excites
- L18:66 weakly excites
- L8:2252 weakly excites
- L11:223 weakly suppresses
- L16:141 weakly excites
- L16:257 weakly suppresses
- L4:1363 weakly excites
- L6:1267 weakly excites
- L1:2920 weakly suppresses
- L4:92 weakly excites
- L6:2783 weakly suppresses
- L8:84 weakly suppresses
- L7:771 weakly excites
- L5:948 weakly excites digits inside years like 19xx
- L4:342 weakly excites
Feeds later layers
- L27:834 suppresses
- L26:1517 suppresses
- L27:1682 weakly suppresses
- L27:285 weakly suppresses
- L26:941 weakly suppresses
- L26:2745 weakly excites
- L26:739 weakly suppresses
- L27:2382 weakly excites
- L27:2763 weakly excites
- L27:1001 weakly excites
- L26:2374 weakly suppresses
- L27:2295 weakly excites
- L26:2427 weakly suppresses
- L27:1060 weakly excites
- L27:2361 weakly suppresses
- L26:2479 weakly suppresses
- L26:2593 weakly excites
- L26:1055 weakly excites
- L26:329 weakly excites
- L26:1158 weakly suppresses
- L27:2397 weakly excites
- L26:2350 weakly excites
- L26:1997 weakly excites
- L27:923 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
- tridge
- 靡
- 子
- че
- maxY
- isons
- .setMax
- max
- شن
- setMax
Suppresses
- class
- Class
- Class
- class
- classes
- (class
- -Class
- -class
- getClass
- getClass