'which' after 'find out' or 'choose'
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. 1 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 'which' when it introduces a choice, such as 'to find out which', 'choose which' or 'predict which'. Its negative side is weak and tied to 'of' after 'each'. Turning it up did not give an obvious theme in the output.
Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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.
-
cell in the cytoplasm, to the ribosomes. There the information of the mRNAs is used for protein production (translation). The fundamental question was which of the two processes, i.e. transcription or translation, plays the dominant role
-
need to choose whatever sound we want to listen to. There are a lot of noises in the world that we don’t want to hear. We need to choose which sound that good for us. If we can’t choose, we will live our
-
on repeating patterns with shapes, sounds and actions. When we build our patterns we count to see how many shapes are in the core. We are learning to predict which shapes might come next. We are also learning about increasing and decreasing patterns and figuring
-
might have to check your risk for heart and blood vessel problems depend on your age, health, gender, and risk factors. Talk to your doctor to find out which tests are right for you.⏎Other Works Consulted⏎- Expert Panel on Integrated
-
’t be performed, and that is called lactose intolerance. Many types of research show the benefits of almond and soy milk, but the question asked multiple times is which one is better? So, let’s find out by comparing these two kinds of
-
even sure exactly what that entails or means. We should be focusing on what the title suggests, NC History. The standards leave it up to the teacher to choose which events or people they prefer to teach about without a clear, concise plan. This
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 am trying to understand what I can do. I am trying to understand what I can do. I am trying to understand what I can do. I am trying to understand what I can do. I am trying to understand what I can
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
healthy eating as defined by community members themselves.⏎This study includes three of the most deprived electoral wards in Stoke-on-Trent. In each of these areas, environmental factors including proximity of physical activity spaces, greenspace and leisure
-
assistance was required several times a day. For a description of what is included in each of these activities, please see the Glossary.<|endoftext|>Madison is preparing for battle
-
OMEN'S BOARDS ORGANIZED⏎After 1850, women organized mission societies that operated independent from maleled denominational boards. In each of five major Protestant denominations women set up mission boards to channel resources specifically toward women
-
into two major groups. Sections 1 through 7 tell the story of agriculture and farmworkers in our area by decade over the past thirty years.⏎In each of the Centerís three decades, certain ethnic communities have been prominent in the diverse farm
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
- L0:1301 weakly suppresses
- L0:1284 weakly excites
- L0:1224 weakly suppresses
- L0:178 weakly suppresses
- L0:1345 weakly excites
- L0:1931 weakly suppresses
- L0:310 weakly excites
- L0:663 weakly excites
- L0:1058 weakly excites
- L0:2120 weakly suppresses
- L0:1 weakly suppresses
- L0:1459 weakly suppresses
- L0:1470 weakly suppresses
- L0:1717 weakly suppresses
- L0:275 weakly excites
- L0:1167 weakly suppresses
- L0:3040 weakly excites
- L0:588 weakly suppresses
- L0:2849 weakly excites
- L0:2183 weakly suppresses
- L0:2765 weakly excites
- L0:2848 weakly suppresses
- L0:377 weakly suppresses
- L0:382 weakly suppresses
Feeds later layers
- L2:147 excites
- L10:2949 weakly suppresses
- L10:167 weakly excites
- L3:58 weakly excites
- L3:2213 weakly suppresses
- L7:1683 weakly excites
- L27:2509 weakly excites
- L18:834 weakly excites
- L5:2546 weakly suppresses
- L2:1259 weakly excites
- L26:2362 weakly excites
- L2:918 weakly suppresses
- L7:2755 weakly excites
- L6:352 weakly excites
- L18:1420 weakly excites
- L14:854 weakly excites
- L19:1842 weakly suppresses
- L21:176 weakly excites
- L7:396 weakly excites
- L5:125 weakly suppresses "of" after all three, both, ten
- L13:1720 weakly suppresses
- L5:2784 weakly suppresses
- L4:285 weakly excites
- L7:883 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
- discipl
- ger
- door
- atcher
- ガー
- imin
- 电器
- doors
- ERING
- 不一样
Suppresses
- ropic
- edImage
- oundation
- pek
- ioc
- edm
- yah
- ém
- 裸
- +n