'official' and government institution 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. 20 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 lights up on words like 'official', 'office' and 'government' when they appear in formal, institutional writing. Turning it up pushed the model into repetitive loops about 'power' and 'control', and turning it down gave off-topic text.
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.
-
, to retain the full citation information.⏎- This article was written and prepared by U.S. Government employees on official time, and is therefore in the public domain.⏎- This publication may be available
-
asuring Housing Affordability⏎U.S. Department of Housing and Urban Development⏎The views expressed in this article are those of the author and do not represent the official positions or policies of
-
technical skills are required only in the IT sector, or in office-setting jobs. But that’s not true. Some level of technical knowledge and skills
-
Act itself, found in the Statutes at Large, 80 Stat. 383 stated:⏎"Except as otherwise required by statute, matters of official record shall be made available, in accordance with published rule, to persons properly and
-
, color, sex, national origin, age, and religion. It guarantees equal opportunity for individuals with disabilities in public accommodations, employment, transportation, State and local government services, and telecommunications.⏎US Dept of Justice ADA website⏎California Department of Development
-
be a policy of official discouragement of caste. Many a times, the constitutional courts have asked for the
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 power over the power over [in the power over] [in the power over] [in the power over] [in the power over] [in the power over] [in the power over] [in the power over]
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
—they were forced into servitude by circumstances over which they had no control. A quick check of Jesus’ parables shows the relationships between slaves/s
-
like something that you talk about when dealing with finances. But this system actually describes how the three branches of government work together to ensure that no one person or group has all of the power.⏎When the Constitution of the United States
-
matrices, calculus, binomials, roots, etc.⏎For such sophisticated expressions, most academics, researchers and the like use TeX Equations to have full control over formatting and rendering their equations exactly as they intended
-
and Balances⏎The process of creating the Constitution was a real challenge. Everyone involved had concerns about how the government would be run and what person or group would have power. One person who heavily influenced the idea of checks and balances was the Baron
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
- L17:2465 excites
- L9:1519 excites
- L8:3029 excites
- L8:2673 suppresses
- L18:2581 excites
- L12:898 suppresses
- L19:2396 suppresses
- L12:2471 suppresses
- L12:592 weakly suppresses
- L18:3049 weakly suppresses
- L5:2480 weakly excites
- L18:2198 weakly suppresses
- L18:2570 weakly suppresses
- L18:2510 weakly excites
- L4:3023 weakly excites
- L17:2888 weakly excites
- L18:2053 weakly excites
- L18:1649 weakly suppresses
- L12:2953 weakly suppresses
- L18:2633 weakly suppresses
- L15:1644 weakly excites
- L17:1485 weakly suppresses
- L8:2699 weakly excites
- L18:1286 weakly suppresses
Feeds later layers
- L24:2746 suppresses
- L25:1872 suppresses
- L25:2403 suppresses
- L27:1416 suppresses
- L25:2072 suppresses
- L22:2290 suppresses
- L24:2697 suppresses
- L21:2448 suppresses "a" after give or get verbs
- L21:2804 suppresses
- L22:2861 suppresses
- L24:1250 suppresses mixed: 'New' in place names, square roots
- L23:2368 suppresses
- L22:2074 weakly suppresses
- L24:2954 weakly suppresses
- L27:1464 weakly suppresses
- L27:1468 weakly suppresses
- L22:2752 weakly suppresses
- L22:1752 weakly suppresses
- L22:1104 weakly suppresses
- L27:2501 weakly suppresses
- L26:2515 weakly suppresses
- L21:2156 weakly excites
- L21:1000 weakly excites
- L22:1158 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
- imations
- imen
- iments
- Stranger
- iment
- ilm
- ressed
- ем
- .netbeans
- ethoven
Suppresses
- control
- Control
- 控制
- CONTROL
- /control
- power
- control
- controle
- 掌控
- 權