colon before a list or answer
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. 16 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 a colon at the end of a line, right before a list, a set of reasons, or a question and answer begins. Turning it up mostly produced rambling or off-topic text, so the link to the trigger is only a guess.
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.
-
. Take a look below at some of the most common questions that we hear from educators and the solution to solve them:⏎"Where do I start?"⏎You are an educator who either purchased your own classroom
-
are several key reasons why copper stills are highly valued:⏎Excellent Thermal Conductivity of Copper:⏎Copper possesses outstanding thermal conductivity, surpassing
-
that is likely caused by many factors. Some of these include:⏎The Brain May Be Cause of Nausea⏎Copyright © Nucleus Medical
-
know, here's some things to check:⏎Its going to vary by manufacturer, window type, and age, but some basic
-
including:⏎· Species distribution and abundances⏎· Species movements and life stages⏎· E
-
deconstruct, and comprehend music. This helpful guide will give you a great grasp of:⏎- Note value and counting notes⏎- Treble and bass clefs⏎-
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: 2023 11 11 11 11 11 11 11 11 11 11
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
a 1 or 2 doses depends on vaccination history⏎as per Advisory Committee on Immunization Practices annual recommendations on⏎prevention and control of influenza with vaccines.⏎“-” indicates information is not applicable⏎In children, the needle size may
-
need more Conjugate Acids and Bases practice, you can also practice Conjugate Acids and Bases practice problems.⏎What professor is this problem relevant for?⏎Based on our data, we think this problem is relevant for Professor Finn
-
wide range of data sources, as some sources are complementary and their combined use can result in the production of more detailed and comprehensive statistics.⏎13.15. The integration of different sources and the reuse of surveys for several purposes could also reduce
-
is made from sand, soda ash and limestone. Other kinds of glass can contain a variety of other ingredients, which could cause problems in a glass container melting furnace.⏎
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
- L14:238 excites
- L2:75 excites
- L11:2789 suppresses
- L9:25 excites
- L1:1 excites
- L2:126 suppresses
- L6:599 excites
- L9:632 excites
- L2:1494 excites
- L9:159 weakly suppresses
- L14:564 weakly excites
- L14:116 weakly excites
- L1:1029 weakly suppresses
- L14:1027 weakly excites
- L8:32 weakly excites
- L5:158 weakly excites
- L15:2852 weakly excites
- L11:1579 weakly excites
- L2:70 weakly suppresses
- L5:3 weakly suppresses
- L0:1301 weakly excites
- L15:603 weakly suppresses
- L0:80 weakly suppresses
- L1:1340 weakly excites
Feeds later layers
- L20:699 suppresses
- L17:50 excites
- L17:68 excites
- L22:213 suppresses
- L18:1075 suppresses
- L18:2501 weakly suppresses
- L24:531 weakly suppresses
- L20:1705 weakly suppresses
- L17:644 weakly suppresses
- L24:572 weakly suppresses
- L21:412 weakly suppresses
- L19:2591 weakly suppresses
- L18:1179 weakly suppresses
- L22:2160 weakly suppresses
- L24:20 weakly excites
- L25:1529 weakly suppresses
- L17:496 weakly excites
- L19:2144 weakly suppresses
- L18:1185 weakly excites
- L17:2155 weakly suppresses
- L17:426 weakly excites
- L17:336 weakly suppresses
- L17:1281 weakly suppresses
- L27:2486 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
- ني
- ’t
- mes
- ت
- :index
- DRAW
- fluence
- 링
- enced
- SAY
Suppresses
- ...";⏎
- ;"⏎
- ()"⏎
- ..."⏎⏎
- .";⏎
- =";⏎
- !";⏎
- ."; ⏎
- ..."⏎
- ".⏎