commas inside long lists
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. 12 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 the comma or separator between items in a long list, such as a run of numbers, countries, or vegetables. Its strongest reactions are at the list separators, and turning it up did not make the model write lists, so this is a weak result.
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.
-
, 32, 33, 34, 35, 36, 37, 39, 49, 51, 6
-
, cauliflower, cranberries, blueberries, apples, beets, ginger, garlic, Brussel sprouts etc. are good during this seasons.⏎Here
-
IC Clearinghouse on Literacy Education)⏎Descriptors: Adult Literacy, Classroom Techniques, Curriculum Development, Daily Living Skills, Employment Potential, English (Second Language), Grammar, Health Education, Health Services, Individualized Instruction, Instructional Materials,
-
day, because you need heat and direct sunlight to cook the s’mores. You will need graham crackers, marshmallows, chocolate bars, aluminum foil, plastic wrap, tape, glue sticks, and cardboard boxes with lids. TIP
-
ghanistan, Angola, Argentina, Bahrain, Belarus, Belgium, Burkina Faso, Burundi, Cambodia, Cameroon, Colombia, Costa Rica, Czech Republic, Denmark, Ecuador,
-
Koothu; Vethiyal Koothu-Pothuviyal Koothu; Vari Koothu - Vari Santhi Koothu; Aariya Koothu-Tamil<|endoftext|>- System scaling relations⏎System scaling
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 been working on the day, and I have to end my day, but I have to end my day, and then I have to end my day, and I have to end my day, and I have to end my day
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
ESV⏎• Luke 9:58 NASB⏎• Luke 9:58 KJV⏎• Luke 9:58 Commentaries⏎• Luke 9:5
-
: Kâla (Time), Durdarsha (Hard-to-be-seen), Durdarshakâla (Hard-to-be-seen-time), Mâtṛceta (Mother-child), Pitṛceta (Father-child
-
1||Feast of Sacrifice|⏎|Saturday||September 02||Eid ul-Azha Day 2||Feast of Sacrifice|⏎|Saturday||September 30||Ashura||10th day of
-
aki Downs||274|⏎|Kowai River, North Branch||274|⏎|Kowai River, South Branch||274|⏎|Lower Warpara Gorge||278|⏎|Other
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
- L6:2740 suppresses
- L11:567 weakly excites
- L9:84 weakly excites
- L9:1334 weakly suppresses
- L8:131 weakly suppresses
- L1:2680 weakly excites
- L7:333 weakly excites
- L9:710 weakly excites
- L6:728 weakly excites
- L11:2735 weakly suppresses
- L11:876 weakly excites 'other' after 'among' in lists
- L6:913 weakly excites
- L10:2828 weakly suppresses
- L9:594 weakly excites
- L11:1082 weakly excites
- L11:219 weakly suppresses
- L11:64 weakly excites
- L4:2735 weakly excites
- L5:2039 weakly excites
- L10:2949 weakly excites
- L10:1830 weakly suppresses
- L2:1004 weakly excites
- L1:3025 weakly excites
- L8:2576 weakly suppresses
Feeds later layers
- L13:2525 excites
- L13:1016 suppresses
- L15:1912 weakly suppresses
- L15:2199 weakly excites
- L18:946 weakly excites
- L17:534 weakly suppresses comma before a list continues
- L13:372 weakly excites
- L16:473 weakly excites
- L15:2872 weakly suppresses
- L17:387 weakly suppresses
- L15:3048 weakly excites
- L17:1518 weakly suppresses
- L16:1109 weakly suppresses
- L21:937 weakly suppresses
- L13:1187 weakly suppresses
- L18:1352 weakly excites
- L16:660 weakly excites
- L15:2460 weakly excites
- L23:935 weakly excites
- L15:2824 weakly suppresses
- L18:2468 weakly suppresses
- L19:2494 weakly excites
- L13:732 weakly suppresses
- L20:2354 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
- wich
- pressive
- iju
- ception
- ycl
- abra
- 因为它
- etur
- cludes
- ball
Suppresses
- 了自己的
- irsch
- Rein
- ascular
- index
- object
- Fall
- argc
- Railroad
- fall