coercion, abuse and harassment
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. 11 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 words tied to force, abuse and victimisation, such as 'coercion' in a definition of human trafficking or 'harassment' in a bullying story. It is weakly negative on words about business deals and loans. Turning it up gave text that drifted toward distress, but the link is loose.
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.
-
growing criminal enterprises in the world.⏎If you’re not familiar with it, U.S. law defines human trafficking as “the use of force, fraud, or coercion to compel a person into commercial sex acts or labor against their will,” and it
-
- In search of their lost ones.<|endoftext|>The Verbal Reasoning section of the Graduate Record Examination (GRE) challenges a student’s reading comprehension, vocabulary, and sentence completion skills
-
all that is required to ensure each child receives the best education, knowing success in learning cannot be achieved in the same way for every child. They support teachers by providing a map of progression from Graduate through Proficient and Highly Accomplished to Lead
-
not drink so much wine that you get drunk, and do not let drinking become a habit.⏎Luke 6:31 (GNBDK)⏎Do for others just what you want them to do for you.⏎To answer the original question,
-
location can result from an accident or from forceful muscle contractions. A shoulder dislocation can also happen when an individual tries to pull themselves up by holding onto something with one hand while lifting their other arm.⏎The most common type of injury that
-
show-up. He or she can be required to speak, stand in a particular pose, and wear appropriate clothing.⏎Suspects can invoke the Sixth Amendment right to counsel during lineups and showups held after they have been arraigned or indicted.
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 2000 car, and a 2000 car, and a 2000 car, and a 2000 car, and a 2000 car, and a
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
also home to a species of ant that live in the plants' hollow thorns. In a classic example of mutualism, the ants protect the plant in return for shelter and food, said Meehan, who conducted the research while at Vill
-
Unit in Wapato kicked into high gear in 2005 when ARS entered into a research agreement with the California Cherry Advisory Board (CCAB). The collaboration makes sense since Washington
-
believe it or not, the global bond market is more than double the stock market.⏎When you purchase a bond from the bond issuer, you are essentially making a loan to the bond issuer. As the bond price is the amount of money investors pay
-
tasks. Redefinition is "when students connect to a classroom across the world where they would each write a narrative of the same historical event using the chat and comment section to discuss the differences, and they use the voice comments to discuss 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
- L10:1624 suppresses
- L10:1995 suppresses formal nouns after verbs of answering or committing
- L10:204 excites
- L10:508 suppresses
- L6:2131 excites
- L10:2004 excites
- L9:1809 excites
- L10:1928 excites
- L10:1347 excites
- L9:2636 excites unclear: camping and money words
- L6:412 suppresses
- L10:1430 suppresses
- L7:1959 excites
- L9:2922 excites
- L10:2726 excites
- L9:2987 suppresses
- L9:782 weakly excites
- L10:2788 weakly excites
- L8:1862 weakly suppresses
- L10:3050 weakly excites
- L7:2351 weakly suppresses
- L10:2785 weakly suppresses
- L4:2185 weakly excites
- L6:1632 weakly suppresses
Feeds later layers
- L17:3040 suppresses
- L12:931 suppresses
- L12:2994 suppresses
- L20:2536 suppresses words for friend or colleague
- L14:2188 suppresses
- L17:1162 suppresses
- L12:1907 suppresses
- L12:2634 suppresses
- L20:2940 suppresses
- L19:1161 suppresses
- L14:1654 suppresses
- L15:1649 suppresses
- L12:597 excites
- L13:802 excites
- L13:2397 excites
- L22:1822 suppresses
- L12:536 suppresses
- L20:1794 suppresses
- L12:1630 suppresses
- L16:2222 suppresses
- L12:2739 suppresses
- L22:1839 suppresses
- L13:2116 suppresses
- L17:1986 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
- parsers
- 开辟
- 觉悟
- annis
- ift
- efforts
- -parse
- PROFILE
- osph
- ifest
Suppresses
- 给了我
- rous
- 对方
- 关系
- ROUND
- hest
- hôn
- rematch
- ymb
- iciel