the number 5 in a numbered list
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. 21 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 reacts to the number 5 when it starts an item in a numbered list, after items 1 to 4. Turning it up mostly made the model's text wander into unrelated topics, so the list pattern is the clearest part.
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.
-
Electrical Engineering Basics⏎2 – DC Circuit Analysis and Basic Electronic Devices⏎3 – Alternating Current (AC) Fundamentals⏎4 – DC and AC Power⏎5 – Power Factor⏎6 – Demand, Load Factor, Service Factor and Electrical Power
-
tell all about the visit of the angel, and that she was to become the Mother of Christ?⏎4. What virtues especially appear in Mary's conduct here?⏎5. Compare the Annunciation to the fall.⏎6. Who was Nestorius
-
⏎3. Obstacles to the Protection of Wildlife Corridors⏎4. Solutions: Managing for People and Wildlife – The Mission of Wildlife Corridors⏎5. Funding the Future⏎What is a Wildlife Corridor?⏎Biologists have long
-
golden age INC This is far from the 1985-98 Rajiv Gandhi and Rao.⏎4. Major role in uprising the country participation⏎5. Role in revolution⏎Party plays an important role in the economic base of the
-
Pius X that Mary was the New Eve on Calvary.⏎4. Does Pope Benedict XV refer to Mary's work in the objective or the subjective redemption?⏎5. Compare the words of Cardinal Pacelli (in Appendix III B)
-
. What is the relation of external action to love? What is the "heresy of action"?⏎4. What is the relation of love to emotion?⏎5. How can one use his knowledge as a means to increase in the love of
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 want to be the one to be the one to be the one to be the one to be the one to be the one. The more I think of this, the more I know that he is the one that is the one that
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
possible to help the beginners to an understanding of what it is better not to write. For the rest I hope the book explains itself.⏎by Mark Twain - Project Gutenberg⏎This book is a collection of essays on various subjects by America's most
-
: Android, General technology, Google, Media, Services, Augmented reality, eyeglasses, feature, featured, glasses, Google, Google Plus, Google Project Gass, google virtual reality glasses, New York Times, Project Glass, technology
-
percent of its normal value, a depression that only decayed once the pole's location stabilized.<|endoftext|>- Project plans⏎- Project activities⏎- Legislation and standards⏎- Industry context⏎Last
-
un”) is another name for the reply system on an email server.⏎Why are Linux services called daemons?⏎The term was coined by the programmers of the MIT Project MAC. they took the name of maxwell's demon, an imaginary being from
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
- L20:217 excites
- L2:1774 excites
- L20:2681 suppresses
- L19:996 suppresses the digit 5 inside numbers
- L16:37 excites
- L19:118 excites
- L18:924 excites
- L10:1052 suppresses
- L20:221 suppresses
- L4:646 excites
- L16:112 excites
- L18:112 suppresses 'over' meaning time or duration
- L17:278 suppresses 'as' in 'as well as' and 'as much as'
- L19:390 weakly excites
- L16:134 weakly excites
- L16:142 weakly excites
- L2:1699 weakly suppresses
- L19:831 weakly suppresses
- L18:440 weakly suppresses
- L16:39 weakly suppresses
- L6:944 weakly excites
- L18:778 weakly suppresses
- L16:164 weakly excites
- L16:673 weakly excites
Feeds later layers
- L25:535 excites commas inside number lists
- L26:319 excites
- L24:124 excites dollar sign before a money amount
- L27:1147 weakly suppresses
- L24:2421 weakly excites
- L27:325 weakly suppresses
- L22:1780 weakly excites
- L26:1486 weakly excites
- L26:398 weakly excites
- L26:1908 weakly suppresses
- L26:2391 weakly excites
- L26:2831 weakly excites
- L26:2133 weakly excites
- L22:953 weakly suppresses
- L26:389 weakly excites
- L26:2511 weakly excites
- L24:1676 weakly suppresses zeros in large round numbers
- L26:1696 weakly excites
- L23:900 weakly suppresses
- L23:1099 weakly excites
- L26:1610 weakly excites
- L23:1577 weakly excites
- L26:920 weakly suppresses
- L24:557 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
- emark
- etics
- csi
- Riot
- lates
- ograf
- kah
- emarks
- НЕ
- mia
Suppresses
- iles
- or
- ural
- IDL
- Firestore
- Gutenberg
- ors
- built
- orris
- ers