'a' before 'few' or 'couple'
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. 25 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 'a' when it is followed by a small amount like 'a few', 'a couple' or 'a moment'. Turning it down made the model repeat 'a' endlessly, and turning it up made the text wander off topic.
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 facilitate the shedding of the uterine lining. Uterine polyps can vary in size, with the smallest being only a few millimeters wide. The growths mostly occur in women who are going through
-
a thorough scan on your system. Before you do that, though, take a few extra precautions.⏎Is your computer part of a network?⏎If so, you
-
area of electronics. A myriad of modern items ranging from cell phones to digital watches, elevators to washing machines contain microprocessors. It is incredible that, just a few decades ago, the microprocessor did not even exist, and yet today it
-
and choose your view.⏎When the system has enough activity and grade data to display, you'll receive an alert in your activity stream. You might need to wait a couple of days as a certain level of student engagement is needed to trigger this alert
-
're making pancakes. You pour 100 grams of flour, a little less than a cup, into the bowl. You add 1 gram, about a quarter of a teaspoon, of salt. After all the other ingredients, you mix
-
you would be willing to participate."⏎After a moment's hesitation, the confederate agreed and started to get up. The experimenter then said "Just a moment," and turned to the subject and repeated the same request. If the subject
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: A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
example, try walking for 15 minutes at an average pace. If you feel this is comfortable, you can increase your time or intensity (only adjust one at a time to ensure that your body is ready for the challenge). Alternately,
-
around Europe, since France was the trend setter then. The new invention was most often called "steps" because you had to step on them, usually one foot at a time, but a set of steps was named after their inventor, Stairs
-
by testing them for anti-HIV activity. In test tube experiments called “assays,” new compounds are added one at a time to cell cultures or actual human cells grown in a laboratory. The goal
-
plants grow best when you develop an affinity for them, and this comes from close observation of the plants plus a large application of common sense. Just take one step at a time. Master the fundamentals before you succumb to the excitement of special techniques
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
- L19:2358 suppresses "any of" before "these" or "the"
- L23:2809 suppresses
- L17:853 suppresses
- L20:1279 suppresses
- L17:2617 suppresses
- L22:1365 suppresses
- L20:1692 excites
- L23:2900 weakly suppresses
- L22:2725 weakly excites
- L4:1625 weakly suppresses
- L24:1024 weakly suppresses
- L18:912 weakly suppresses
- L20:3069 weakly suppresses
- L12:497 weakly excites
- L13:1305 weakly suppresses
- L24:2538 weakly suppresses
- L21:395 weakly excites
- L15:869 weakly suppresses
- L6:1895 weakly suppresses
- L21:2353 weakly suppresses
- L19:233 weakly suppresses
- L14:2832 weakly suppresses
- L22:1033 weakly excites
- L21:708 weakly suppresses
Feeds later layers
- L27:797 strongly suppresses
- L27:231 strongly excites
- L27:663 strongly suppresses
- L27:709 strongly suppresses
- L26:484 suppresses
- L26:213 excites
- L27:173 excites
- L26:411 excites
- L27:2396 suppresses
- L26:829 suppresses
- L26:1102 weakly excites fibromyalgia and hour counts
- L27:108 weakly excites
- L27:854 weakly suppresses
- L27:1412 weakly excites
- L26:2417 weakly excites
- L26:3024 weakly suppresses
- L27:1292 weakly suppresses
- L26:2458 weakly excites
- L27:330 weakly suppresses
- L27:580 weakly excites
- L26:1652 weakly suppresses
- L27:1635 weakly excites
- L26:991 weakly excites
- L26:2327 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
- ős
- empo
- eria
- erie
- ondheim
- ihan
- utt
- руд
- __/
- IMER
Suppresses
- a
- -a
- →a
- )a
- *a
- ,a
- _a
- a
- ;a
- ’a