closing parenthesis after a number or unit
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. 10 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 fires on a closing bracket that ends a short phrase, usually right after a number, unit or a word like 'serous'. Turning it up did not produce a clear theme in the 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.
-
0 mph)<|endoftext|>Today I wondered which NT books relied more heavily on the Septuagint
-
North American) family.⏎Where Do Anchovies Come From?⏎Anchovies are saltwater for
-
15 cm) in⏎tangled, bushy clumps. Axes extend through entire length of
-
rolled oats) to 12 hours (oxtail). Foods mostly cook for more or
-
of effectiveness) each year produces over $350,000 more income for his
-
cemia) occurs when there is too little glucose (sugar) in your child’s blood
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 think that, but not all, and I think that, but not all, and I think that, but not all, and I don't. I think that, and I don't, but I don't. I
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
school gardens and shows the value of these small urban plots to educators, students, and the communities served by these gardening projects. Highlights small gardens by region, recounting the history of each garden project, student experiences there, and the pedagogical
-
perfect companion for study and revision for nursing and healthcare students from the publishers of the market-leading at a Glance series. This easy-to-read, accessible guide brings together all the key principles of caring for patients with medical and surgical needs. Highly visual
-
, a goal that finally was achieved when the Nineteenth Amendment passed in 1920. Three years later, at the 75th anniversary of the Seneca Falls Convention, Alice Paul proposed passage of an Equal Rights Amendment to
-
C.⏎PUBLISHER: Elsevier - Health Sciences Division⏎Also available at Amazon.com⏎Calculate with Confidence provides a clear consistent format with a step-by-step approach to the calculation and administration of drug dosages. It covers the ratio and
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
- L7:879 excites
- L6:536 weakly excites
- L2:618 weakly excites
- L9:469 weakly excites
- L2:64 weakly excites
- L4:1038 weakly suppresses
- L5:1732 weakly excites
- L4:2076 weakly excites
- L5:2061 weakly suppresses
- L7:1003 weakly excites
- L0:870 weakly excites
- L5:1956 weakly excites firsthand accounts and testimony
- L9:1425 weakly suppresses
- L3:1444 weakly excites
- L6:142 weakly suppresses
- L7:926 weakly excites
- L8:1185 weakly excites
- L0:2227 weakly excites
- L5:1505 weakly suppresses
- L9:1413 weakly suppresses
- L5:1479 weakly suppresses
- L9:894 weakly suppresses
- L1:2191 weakly excites
- L8:582 weakly excites
Feeds later layers
- L17:506 excites
- L19:490 excites
- L11:2310 excites opening parenthesis in formulas and math
- L11:3059 excites
- L14:1195 excites closing bracket ending an aside
- L13:2512 weakly excites closing parenthesis ending an aside
- L15:324 weakly excites
- L11:547 weakly excites
- L11:2191 weakly excites unclear: brackets and parentheses in text
- L12:2375 weakly excites
- L17:1708 weakly excites
- L15:2624 weakly excites
- L13:407 weakly suppresses
- L20:273 weakly suppresses
- L14:2130 weakly excites
- L12:575 weakly excites
- L17:1824 weakly excites
- L21:1340 weakly excites
- L19:260 weakly excites
- L16:289 weakly excites
- L16:2449 weakly excites
- L20:1065 weakly excites
- L13:1698 weakly excites
- L20:2211 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
- 商报
- akra
- ackers
- ighborhood
- touches
- ynet
- ネ
- ーム
- truncate
- nish
Suppresses
- ligt
- .fhir
- isp
- ashion
- 徭
- ово
- :initComponents
- bler
- 发展
- stood