end of web page disclaimer
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. 18 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 is strongest at the invisible end-of-page marker that comes after a medical or health disclaimer, such as 'consult your doctor'. Its negative side is weak and mostly about numbers. Steering did not produce a clear health theme, just rambling or repeated text.
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.
-
direct supervision of a qualified health practitioner such as a naturopathic physician.<|endoftext|>In 1939 the Harvard sociologist Edward Hartshorne gathered eyewitness
-
intended to be a substitute for professional medical advice, diagnosis, or treatment. Always seek the advice of qualified health providers with questions you may have regarding medical conditions.<|endoftext|>The Amendment Process and the African American Fight for Representation and Equality⏎Black History Month
-
given as suggestions only. We recommend you also consult your healthcare provider, and urge you to contact them immediately if your question is urgent or about a medical condition.<|endoftext|>The Joint United Nations Programme on HIV/AIDS (UNAIDS) confirmed that in
-
treat patients who only have inflammation in the stomach. Triple therapy is not right for every child. Ask your doctor if it is the right treatment for your child.<|endoftext|>Driving an embedded car save not just time but it can save lives as well.
-
, each of which is shown in turn below:⏎The distance between any grid point and the rectangular border is one unit of measurement (equal to the length of the<|endoftext|>Shree Ganeshay Namah⏎How to write Anchoring Script in English
-
ions in the region, and are insufficient to offset present rates of relative sea level rise. © 2006 Elsevier Ltd. All rights reserved.<|endoftext|>Are you tired of being sick and tired of being sick and tired? Then,
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: it's a more interesting and more interesting than the materialistic and materialistic stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
13, 81 ce), Roman emperor (79–81), and the conqueror
-
Egypt, El Salvador, Finland, Grenada, Guatemala, Japan, Lao, PDR, Macedonia,
-
1, 1, 2, 3, 5, 8, 1
-
1, 1/4 turn clockwise and then back 2_______.⏎5.1
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
- L17:2635 strongly excites
- L17:2002 strongly excites
- L15:1898 excites
- L17:1527 excites
- L17:1586 excites
- L13:257 excites
- L16:1430 suppresses
- L17:2327 suppresses
- L16:562 suppresses
- L14:978 suppresses
- L16:2077 excites
- L13:1860 excites
- L16:2962 suppresses unclear: weak mid-phrase prepositions
- L17:2895 excites
- L13:2182 suppresses
- L15:2759 suppresses
- L14:2525 suppresses
- L16:2943 excites
- L17:1635 excites
- L14:66 suppresses
- L15:158 suppresses copula after 'probably' or 'may'
- L15:1959 excites
- L16:1313 excites
- L16:631 excites
Feeds later layers
- L19:1567 strongly excites
- L24:174 excites
- L20:759 excites
- L19:355 suppresses
- L20:697 excites
- L19:1673 excites
- L19:401 suppresses
- L21:295 suppresses
- L24:367 excites
- L23:348 excites
- L19:823 suppresses
- L19:2218 excites
- L19:1336 excites
- L20:708 suppresses
- L19:1291 excites
- L21:457 excites
- L19:225 suppresses
- L19:1446 excites
- L22:865 excites
- L19:857 excites
- L20:1039 suppresses
- L21:403 excites
- L19:1789 excites middle of personal names
- L27:141 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
- night
- shore
- night
- edy
- -night
- 夜
- Crypt
- otope
- Night
- Nights
Suppresses
- ')")⏎
- '))⏎
- ĩnh
- >*/⏎
- %@
- chal
- 眷
- +%
- 詳
- ()}}⏎