middle of a fixed multi-word term
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. 13 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 a word in the middle of a fixed name or term, such as "emerald ash borer", "weighted average cost of capital" or "foot-and-mouth disease". The pattern is loose, and steering did not reflect it.
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.
-
. E.g. I had gone to the shop and bought some chips.⏎Other examples include.⏎After she had read it from start to finish, Sophie ripped the letter in half.⏎The frog had jumped from the lilypad to the shore.⏎Tom
-
. People at risk include those whose occupations expose them to blood, such as health-care workers. The vaccine is now commonly given to all newborns.⏎• Hepatitis A vaccine: Hepatitis A is transmitted through contaminated food or water. The Hep
-
, provides goal writing guidance, especially useful for early intervention and children on the autistic spectrum disorder.<|endoftext|>The online WACC Calculator is used to calculate the weighted average cost of capital (WACC).⏎In finance, The weighted average cost of capital,
-
. Have some introduced species caused changes that most everyone would agree have been very harmful? Absolutely. By killing timber trees, gypsy moths and the emerald ash borer have caused, and continue to cause, enormous economic damage to the United
-
oofed animals. In 1981, Chile was officially declared FMD-free without vaccination, after complying with all international health requirements.⏎Foot-and-mouth disease (FMD) affects mainly swine and related wild ruminants and
-
complain to police in Karnataka online?⏎- What does an eagle with two heads mean?⏎- Did the Hittites invade Egypt?⏎- What does the word Hittite mean?⏎- Who were the Hittite gods?⏎- What did 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: I think it's important to make sure that the world is in your thoughts. I think it's a thing of making sure you are always making your thoughts, but I think there is a lot more to it. I think you are
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
41–46); Samuel, the altar boy who became God’s prophet (1 Samuel 3:10–21); David, the shepherd boy who became king (1 Samuel 16); Esther, the teenage queen who
-
Library of Congress) (1890)⏎"To avoid these evils, each gathering hule has its "boot," a rounded hood of fire-clay which surrounds the hole on the inside
-
do you use the word Which?⏎- Who is example sentences?⏎- Why is it called crush?⏎Which means in a sentence?⏎He’s a politician now, which means that bragging and whoring his story is the job description.⏎The ech
-
istic methods uncover sophisticated meanings and monkey dialects⏎The same species of monkeys located in separate geographic regions use their alarm calls differently to warn of approaching predators, a linguistic analysis by a team of scientists reveals. The study, which appears in the journal
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
- L11:1482 excites
- L11:1368 excites
- L12:2778 excites
- L12:1712 excites
- L11:1107 suppresses
- L11:1692 suppresses
- L11:1298 suppresses
- L12:1547 excites
- L12:1074 excites
- L11:722 excites
- L1:853 excites
- L11:888 suppresses
- L9:600 excites
- L12:1359 suppresses
- L11:442 weakly excites
- L9:1981 weakly excites
- L10:826 weakly excites
- L10:542 weakly excites word completing a technical term
- L10:198 weakly suppresses
- L12:114 weakly excites
- L11:284 weakly excites
- L12:685 weakly suppresses
- L2:742 weakly excites
- L12:1683 weakly suppresses
Feeds later layers
- L15:1769 strongly suppresses
- L14:2284 suppresses
- L14:331 suppresses
- L14:1860 excites
- L14:1603 excites
- L15:1502 suppresses
- L14:410 excites
- L16:1021 excites
- L16:1738 excites
- L16:2944 suppresses
- L14:369 suppresses
- L14:372 weakly excites
- L14:97 weakly excites
- L23:708 weakly suppresses
- L16:2438 weakly suppresses
- L14:336 weakly excites
- L14:2995 weakly excites
- L14:1113 weakly suppresses
- L14:644 weakly suppresses
- L15:359 weakly suppresses
- L17:1000 weakly excites
- L15:2044 weakly suppresses
- L14:654 weakly suppresses
- L16:2152 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
- moduleId
- Opens
- 不排除
- OUCH
- broad
- fry
- drifted
- hire
- ']+
- YLE
Suppresses
- �
- 年中国
- 变化
- .DrawString
- istan
- reco
- aling
- 手腕
- 等奖
- velle