"of" in "Department of"
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. 2 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 the word "of" right after "Department", as in "Department of Homeland Security". Turning it up made the writing wander without a clear pattern.
Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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.
-
. Department of Homeland Security's (DHS's) Cybersecurity and Infrastructure Security Agency (C
-
6⏎Department of International Health⏎Objective: To explore how Ghanaian socio-cultural beliefs, household power
-
the public domain cannot be commercially exploited without permission of the copyright⏎owner. Responsibility for any use rests exclusively with the user.⏎Copyright Owner: UC Regents⏎Department of Special Collections, Charles E. Young Research Library, U.C. Los Angeles<|endoftext|>
-
a diagnosis is only made once the symptoms become so prominent that they interfere with our everyday lives."⏎Predrag Petrovic, Consultant, Associate Professor, Psychiatry, Department of Clinical Neuroscience, Karolinska Institutet⏎The study included 18
-
7 made the office an elective position.⏎The Department of Law, with the attorney general as its head, was established by the General Assembly
-
• Teacher Resources • Frequently Asked Questions • Glossary • References • Links to Other Mussel Sites⏎Privacy • FOIA • FirstGov • Contact⏎Department of the Interior • U.S. Fish & Wildlife Service • U.S. Geological Survey
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 a lot of good luck to the stars, and to the sky above me, and to the sky above me. I have a lot of good luck to the stars, and to the sky above me. I have a lot
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
PRODUCING THEATER⏎- CREATING THEATER⏎- THE CREATIVE TEAM⏎- THE CREATIVE ARTISTS⏎- IMPLEMENTING THE DESIGNS⏎- PRODUCTIONG MANAGEMENT / STAGE MANAGEMENT / TECHNICAL D
-
but more indicates an excessive buildup. In addition to bagging lawn clippings, aerating the<|endoftext|>Comments. They are some of the magic sauce of ds1
-
does he depict the sixteen-foot-wide alleys that are clearly platted on the Sanborn maps, although in several cases there appear to be spaces for the undesignated alleys. But, on the whole, he produced a crisp and positive
-
problems and to explore opportunities for positive action, individually and collectively.”⏎Other nations were far more skeptical. France and the United Kingdom, for example, were<|endoftext|>Diseases & Conditions⏎If you trim your toenails too short, particularly on the
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
- L0:1301 suppresses
- L0:1224 suppresses
- L0:178 suppresses
- L1:2596 suppresses
- L1:1493 weakly excites
- L1:1960 weakly excites
- L1:11 weakly excites
- L1:1461 weakly excites
- L1:1464 weakly excites
- L0:2270 weakly suppresses
- L1:491 weakly excites
- L1:187 weakly suppresses
- L1:1555 weakly suppresses
- L1:724 weakly excites
- L1:2318 weakly suppresses
- L0:2248 weakly suppresses
- L0:1137 weakly excites
- L1:2683 weakly excites
- L1:1692 weakly excites
- L1:1583 weakly excites
- L0:282 weakly excites
- L0:2156 weakly excites
- L1:1009 weakly suppresses
- L0:1016 weakly suppresses
Feeds later layers
- L5:2486 weakly excites
- L3:1811 weakly suppresses
- L3:2820 weakly suppresses
- L6:828 weakly suppresses
- L11:1083 weakly excites
- L22:1184 weakly excites
- L25:74 weakly excites
- L15:706 weakly excites
- L3:422 weakly excites
- L3:1623 weakly excites
- L7:856 weakly suppresses
- L3:980 weakly excites
- L7:2409 weakly excites
- L6:1202 weakly excites
- L19:2946 weakly excites
- L9:1635 weakly suppresses
- L12:1972 weakly excites
- L13:2736 weakly excites
- L12:124 weakly excites
- L7:1404 weakly excites
- L6:2796 weakly excites
- L4:2027 weakly excites
- L5:1323 weakly excites
- L4:2623 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
- ]),⏎
- ]**
- atin
- ']),⏎
- switch
- ])**
- []),⏎
- ]),⏎
- edish
- ])),⏎
Suppresses
- ocos
- jas
- 已经是
- rozpoc
- 限り
- apes
- %n
- IZ
- زال
- 已是