'is' after focus or goal nouns
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 'is' when it follows a noun like 'focus', 'goal' or 'main challenge', as in 'The focus is to...'. Turning it up did not produce a clear theme in the model's replies.
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.
-
skills and a “family” of supports and resources.⏎c. Inclusive ECE Coaching is intended for EHS/HS program coaches and supervisors. The focus is to build a Community of Reflective
-
Surgery. His focus is to safely return athletes to the playing field of their choice and to keep them there
-
ey. I am an experienced programmer, having received several accommodations on programming in my formal education as well as tough several programming languages to thousands of students. My goal is
-
ones as the child advances additional skills can be introduced. The focus is first set on the ability of the child to perform simple skills such as eating properly
-
1986. Now he travels around Namibia, providing stargazing and astronomy education through his company, Space Observation Learning Namibia. His goal is to have more young heads in the clouds.⏎"Nobody looks up anymore," John
-
classes to see how the students produced out with the projects. The website gives a nice collection of classroom activities that incorporate reading, writing, and math. Their focus is to make certain that all schools are living up to
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 been thinking about lately: I have I have 4 I have 5 I have 6 I have 7 8 9 1 1 2 1 2 3 4
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
to linguistic errors are influenced by both the effectiveness and efficiency of domain-general conflict monitoring. These relationships were
-
properties. This information is especially important when enhanced-oil-recovery techniques that use expensive fluids such as surfactants, micellar fluids, or polymers are considered. In the last 50 years, many tracer studies have been reported
-
in these schools are assumed to be influenced by religion in their development, including ways of being citizens and
-
riting about those points to see which might make a useful topic. It allows them some room to explore potential topics before having to commit to one, and thus alleviates some pressure
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
- L1:1100 suppresses
- L0:2283 excites
- L1:174 suppresses
- L1:268 excites
- L1:462 excites
- L0:406 excites
- L1:242 suppresses
- L0:632 weakly excites
- L1:700 weakly suppresses
- L1:993 weakly excites
- L1:505 weakly excites
- L1:1360 weakly excites
- L1:1007 weakly excites name pieces inside a multi-part word
- L0:670 weakly excites
- L1:105 weakly excites
- L1:579 weakly suppresses
- L1:239 weakly excites
- L1:1001 weakly excites
- L1:736 weakly suppresses
- L1:175 weakly excites
- L0:816 weakly suppresses
- L0:847 weakly suppresses
- L1:1579 weakly suppresses 'Well' starting a reply or answer
- L0:444 weakly excites
Feeds later layers
- L4:22 strongly suppresses
- L4:476 strongly excites
- L5:395 strongly suppresses
- L6:155 strongly suppresses
- L5:99 strongly excites
- L11:344 strongly excites
- L8:511 strongly suppresses
- L15:304 strongly excites
- L15:318 strongly suppresses
- L10:422 excites
- L5:351 suppresses
- L7:84 excites
- L9:157 suppresses
- L6:769 suppresses
- L10:586 suppresses 'important' and degree words after 'how'
- L16:968 excites
- L11:782 excites
- L16:881 excites
- L4:152 suppresses
- L14:850 suppresses
- L17:2216 suppresses
- L5:460 suppresses
- L8:1149 suppresses
- L10:766 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
- /her
- gì
- disguise
- &q
- ;z
- %;
- Dİ
- 一听
- DEFINE
- <nav
Suppresses
- 也是如此
- 是有
- antan
- burgh
- (){}⏎
- 中有
- ']),⏎
- }()⏎
- 拥
- るのは