the blank spot right before a number
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 fires on the space just before a number, usually right after words like 'of', 'by', 'from', or 'increased', where a figure such as a percentage, age, or year is about to appear. Turning it up produced loose, unrelated text, so the steering does not show this theme.
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.
-
. Fully grown males tend to weigh less than 6 ounces, and females less than 4 ounces. So, how large do chameleons grow? The
-
linked to the U.S. explosion in childhood obesity. One study of Massachusetts schoolchildren found that for each additional sweet drink per day, the odds of obesity increased 60 percent.⏎As a result, many schools have stopped selling pop or artificial
-
moving along the parabola 4y = (x + 2)^2 in such a way that its x-coordinate is increasing at a constant rate of 2 units per second. How fast is the particle's distance to the point (-
-
of heart attacks and 54% of strokes worldwide, the authors write. They note that the risk of heart attack and stroke drops by about half for every 20 mm/Hg reduction in systolic blood pressure (the top number of
-
to generation.⏎The biggest risk factor for developing PD is advancing age. The average age for the onset of PD is 60 years. In addition, 50 percent more men are affected than
-
U.S. Council of Environmental Quality, the cancer risk to people who drink chlorinated water is 93 percent higher than among those whose water does not contain chlorine. The EPA
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 don't know how to start, or how to know if I'm actually trying to write a story." Okay, let's see. I'm a writer, and this is a common challenge. I need to keep going until I
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
a landscape digital photographer. It does not utilize the camera. The professional photographer takes pictures of natural scenes like beaches, hills, or falls. Modern modern technology has actually made it possible for expert landscape photographers to create pictures
-
engaged students into a state of active learning. They are present only physically, with their mind elsewhere. They don’t read the materials, do the homework and don’t ask questions, or engage in discussion. Instead, they just watch or listen
-
Table of Contents⏎- 1 How welding works⏎- 2 Which welding is best for welding pipes?⏎- 3 Tools required for pipe welding⏎- 4 Steps to welding a pipe⏎- 5 Welding safety precautions
-
ellec 40 for the 4-Bit microprocessors i4004 and i4040, Intellec 8 and Intellec 8 mod 80 for the 8-Bit processors.⏎Due
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
- L8:9 excites
- L8:187 suppresses
- L7:83 suppresses
- L8:319 excites
- L8:272 excites
- L1:193 weakly suppresses
- L7:135 weakly suppresses 'of' after 'issue' in citations
- L8:361 weakly suppresses
- L6:123 weakly suppresses
- L3:0 weakly excites
- L8:395 weakly suppresses
- L1:116 weakly excites
- L4:1097 weakly suppresses
- L4:541 weakly suppresses
- L4:515 weakly excites
- L3:757 weakly excites
- L5:664 weakly suppresses
- L4:348 weakly suppresses
- L8:1453 weakly suppresses
- L7:184 weakly excites
- L8:2161 weakly excites
- L2:791 weakly excites
- L8:1821 weakly excites
- L8:1775 weakly suppresses
Feeds later layers
- L10:541 suppresses
- L11:135 weakly suppresses
- L10:42 weakly suppresses
- L10:97 weakly suppresses
- L10:62 weakly suppresses
- L11:56 weakly suppresses
- L13:13 weakly suppresses opening parenthesis after a term
- L19:2757 weakly suppresses
- L10:407 weakly suppresses
- L14:4 weakly suppresses
- L12:704 weakly suppresses
- L10:1090 weakly excites
- L10:497 weakly excites
- L11:6 weakly suppresses
- L13:1358 weakly excites
- L12:887 weakly excites
- L15:371 weakly suppresses
- L12:869 weakly excites
- L11:123 weakly suppresses
- L15:3050 weakly excites
- L18:2642 weakly suppresses
- L27:2673 weakly excites
- L24:1037 weakly suppresses
- L21:2718 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
- reds
- erez
- epad
- erness
- ermal
- eren
- 、「
- ipline
- puts
- oplevel
Suppresses
- 安东尼
- arth
- andid
- .shared
- สรร
- ewidth
- 堑
- @endforeach
- 一年一度
- =settings