word before 'that', 'to' or 'for' continuation
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. 15 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them 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 a word that is about to be followed by a link word like 'that', 'to' or 'for', as in 'the energy that' or 'suitable for'. It is strongest where the sentence needs a continuation to finish a thought. Turning it up did not produce a clear theme in the model's replies.
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.
-
rich avifauna, characteristic of the country’s central highlands<|endoftext|>Luckily, Music Theory for Dummies shows you the fun and easy way to understanding the concepts needed to compose,
-
it to emit a great amount of radiation, mostly at visible wavelengths. The Sun size also contributes to the large amount of radiation.⏎The Sun provides all the energy that drives our climate. Global warming is a gradual increase in the average temperature of the
-
use the terms “earth”, “land”, or “ground” to refer to the same thing. Whilst you might understand soil as just the dirt that you use to pot your plants, it is so much more than that.⏎Soil is the
-
of the most influential men in his own thinking and was even accused of plagiarizing Locke while writing the Declaration of Independence.⏎In Locke we find the arguments and ideals that inspired our founding documents. For example, Locke famously argued that our dignity comes directly
-
When your cells can’t get glucose (a kind of sugar) to burn for energy, they burn fat instead. This leaves behind acids called ketones in your
-
on n-body algorithms rooted in the principles of Newtonian mechanics. The same techniques that predict planetary motions here at home in the solar system also describe the gravitational process that assembles thousands of galaxies into filaments a hundred million light-years long.⏎A
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 have a 30000000000000000000000000000000000000000000
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
ites reproduce rapidly. This allows large populations to develop and damage your tree severely before you even notice the symptoms. On top of that, oak spider mites spread throughout you and your neighbor’s property easily. Once one
-
. It is essential that you understand and comply with this legislation when purchasing an object of historical, architectural or archaeological interest. Buying with confidence will help you to comply with the legislation. Factsheet Legislation provides further information.⏎What international legislation governs the
-
the unit hydrograph.⏎Upon successful completion of this course, the student will be able to:⏎CIVL 1060 is offered as a part of the following programs:⏎If you have a comment or question about this course, please
-
about 10,180,000 square kilometres (3,930,000 sq mi), or 2% of the Earth’s surface (6.
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
- L14:1547 excites
- L14:2163 suppresses
- L13:723 suppresses
- L12:2685 suppresses
- L14:2512 excites
- L12:2151 suppresses
- L13:2238 suppresses
- L12:2850 suppresses
- L14:1281 suppresses
- L14:576 excites
- L12:2694 suppresses unclear: 'purpose' and words near it
- L14:1821 suppresses
- L10:322 suppresses
- L12:522 suppresses
- L14:522 excites
- L10:2412 suppresses
- L13:2493 suppresses
- L12:411 excites
- L12:3004 excites
- L14:329 suppresses
- L11:2037 excites
- L13:135 suppresses document boundary end-of-text token
- L12:2478 suppresses
- L12:751 suppresses
Feeds later layers
- L16:1266 excites
- L16:347 excites
- L16:1540 suppresses
- L16:814 suppresses
- L18:299 suppresses
- L16:1008 suppresses
- L17:112 suppresses
- L16:394 excites
- L19:522 suppresses
- L17:1567 suppresses
- L16:316 excites
- L16:351 suppresses
- L16:2289 suppresses 'or' linking two alternatives
- L18:360 suppresses
- L17:168 excites
- L20:289 weakly excites
- L16:288 weakly excites
- L16:1229 weakly suppresses
- L18:160 weakly excites
- L16:1350 weakly suppresses
- L16:1026 weakly suppresses
- L19:2151 weakly suppresses
- L17:2764 weakly excites
- L16:291 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
- Cycl
- eus
- bp
- mount
- actic
- ANCH
- arges
- cycl
- caves
- ROP
Suppresses
- \Id
- estureRecognizer
- (feed
- LinkId
- .writeValue
- setSearch
- newName
- ieties
- 娣
- ->{_