unclear: weak, articles before 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. 10 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 gives only weak responses, mostly on small words like 'the', 'a' or 'some' that come just before a noun in the middle of a sentence. It does not clearly change the writing when turned up or down.
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.
-
organize and prioritize the goals to be accomplished.⏎Popular methods of time management practiced in the 21
-
some extra time to make a decision. It now hopes to make a launch decision on Monday morning.⏎“We have
-
One of the most important requirements is to sample where irradiation and oxidation are minimized, either by
-
One of the items is coffee, but you can’t think of the other one to save your
-
. Here are some great examples in a musical format.⏎We will be looking at numbers next time then
-
first of the series of articles I introduced last week about the nutritional habits of various unique groups
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: the first, the second, or, or, or, or, or, or, or, or, or, or, or, or, or, or, or, or, or, or, or, or, or,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Academy of Sciences, enlarges a growing body of work linking triclosan to human and environmental health issues. In “test tube” experiments, triclosan impairs the ability of isolated heart muscle cells and skeletal muscle fibers to contract
-
use in Europe until silk mesh was more available for trade from the east and a profitable outlet for the medium discovered.⏎Screen printing was first patented in England by Samuel Simon in 1907. It was originally used as a popular method to
-
the best available top probiotic foods.⏎Probiotic Drinks List⏎Probiotic drinks include:-⏎– kefir (fermented milk)⏎– water ke
-
tumors in the brain, including:⏎Herniation of the brain can also be caused by other factors that lead to increased pressure inside the skull, including:⏎Brain herniation can occur:⏎Patients with a brain herniation have:⏎A brain and nervous
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
- L9:15 strongly suppresses
- L5:25 strongly suppresses
- L5:36 strongly excites
- L8:7 suppresses
- L6:7 excites
- L0:18 excites
- L8:10 suppresses
- L0:2331 suppresses
- L8:3 suppresses
- L7:122 suppresses
- L8:4 suppresses
- L8:13 excites
- L4:23 suppresses
- L8:16 excites
- L6:16 suppresses
- L5:4 suppresses
- L7:63 suppresses
- L6:13 suppresses
- L9:5 suppresses
- L8:14 suppresses
- L5:50 suppresses
- L3:59 suppresses
- L6:18 suppresses
- L7:62 suppresses
Feeds later layers
- L12:2159 strongly suppresses
- L11:1353 strongly suppresses end-of-document token after a web snippet
- L11:48 strongly suppresses
- L12:1215 strongly suppresses
- L11:957 strongly excites
- L12:1499 strongly suppresses
- L12:396 strongly excites
- L11:34 strongly excites
- L12:640 strongly suppresses
- L16:192 strongly suppresses
- L11:9 strongly excites
- L19:1247 suppresses
- L15:2763 suppresses
- L12:1079 suppresses
- L13:654 suppresses
- L11:70 suppresses
- L18:596 suppresses
- L13:105 excites
- L15:514 excites
- L13:1044 suppresses
- L19:1299 suppresses
- L15:1278 suppresses
- L12:623 excites
- L14:68 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
- utely
- лист
- 大卫
- �件
- atted
- пат
- cars
- atra
- attle
- departamento
Suppresses
- …⏎
- ^\
- …⏎⏎
- eros
- Slo
- OOM
- ypo
- 错
- …"⏎⏎
- GO