'case' after 'in this'
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. 9 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 lights up on the word 'case' in phrases like 'in this case', and on words that point back to the current example, such as 'here'. Turning it up made the model write about examples and the question at hand, though loosely.
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.
-
you plural|⏎A direct object is a noun following the verb that answers the questions what? or whom? Remember a pronoun replaces a noun, and in this case the noun is a direct object. Verbs that have a direct object are called
-
Hawaii (figure 2). When hydroacoustic activity has been of sufficient amplitude and duration, it has often been inferred to suggest submarine eruptions. In this case the signals were relatively weak and of modest duration, indicating seismic swarms without a
-
to the mites. The upshot is that a rash results.⏎So there are two ways to approach the problem: treat the underlying disease — CLL in your case
-
recently passed separate laws in 2001 that<|endoftext|>Pulse Width Modulation (or PWM) is a technique for controlling power. We also use it here to control the brightness of each of the LEDs. … Roughly every
-
When we judge an action as morally right or wrong, we rely on our capacity to infer the actor's mental states (e.g., beliefs, intentions). Here, we test the hypothesis that the right
-
in Tucson,⏎and lead author of the study, which appears in today's Nature.⏎"Normally, you have the greatest thickness underneath the tallest mountains⏎in this case
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 can be an animal of the year." So, how many times is the word used in a year?" Let me explain the word in a year, and what is its value in a year." So, how many times is the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
is the most appropriate management?⏎vitamin D and calcium levels⏎MRI of the pelvis⏎long leg cast⏎aspiration of the knee⏎A 3-year-old male presents with inability to bear weight on his right leg for the past
-
answers is unusual, for students scoring the way John and Barbara scored. If it is unusual, how unusual is it? If it would only occur one time in a million (or even less often) by chance alone, the instructor may conclude that
-
problems: one each of addition, subtraction, multiplication, division, fractions, and percents. I stressed "grade-appropriate" because single digit addition, for example, is not something most 5th graders need to practice. My expectations
-
seems appropriate to have some humility around this energy transition,” said co-author David Rapson, an associate professor at the UC Davis Economics Department. “Approaches that leave multiple technology pathways open are
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
- L7:726 suppresses
- L8:401 suppresses
- L7:34 excites
- L8:355 excites
- L7:2291 suppresses
- L8:543 weakly excites "Since" starting a reason clause
- L7:318 weakly suppresses
- L7:267 weakly suppresses
- L2:1385 weakly suppresses
- L8:824 weakly suppresses
- L8:374 weakly suppresses
- L5:178 weakly suppresses
- L8:330 weakly excites
- L5:396 weakly suppresses
- L1:2372 weakly suppresses
- L8:207 weakly suppresses
- L3:487 weakly suppresses
- L7:1106 weakly suppresses
- L8:1884 weakly suppresses
- L8:606 weakly excites
- L6:2135 weakly suppresses
- L7:1418 weakly suppresses
- L1:1395 weakly suppresses
- L2:1646 weakly suppresses
Feeds later layers
- L12:2575 excites
- L12:976 suppresses
- L12:1602 excites
- L10:805 suppresses
- L12:1002 excites
- L16:2499 suppresses
- L11:549 weakly suppresses
- L15:1221 weakly excites
- L22:1854 weakly excites
- L14:2718 weakly excites
- L11:78 weakly suppresses
- L11:2619 weakly suppresses
- L15:86 weakly excites
- L11:54 weakly excites
- L12:2942 weakly excites
- L13:669 weakly excites
- L15:753 weakly suppresses
- L11:994 weakly suppresses
- L10:1036 weakly excites
- L13:320 weakly excites
- L10:996 weakly excites
- L10:961 weakly suppresses
- L14:2848 weakly excites
- L11:985 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
- case
- cased
- case
- onaut
- ocoa
- isma
- .case
- -task
- itate
- хот
Suppresses
- }*/⏎
- _under
- 綦
- Portugal
- UNDER
- цип
- }*/⏎
- 又被
- vej
- 韬