opening parenthesis mid-sentence
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 reacts to an opening bracket '(' that starts an aside or a short abbreviation, like '(USDA)' or '(for a long time, at least)'. The same bracket can push the neuron either way, and steering did not produce a clear theme.
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.
-
by itself. Therefore, a number of the more popular study questions include (but are not limited to) whether there are connections between colors and learning, whether
-
ght” in 1993, he became one of the first to author the word aight in published text. He’s even glossed it “(Alright)” for everyone who hadn’t caught on yet.⏎We learn from the O
-
. (This also happens to be the opinion of the FDA, public health officials, and
-
. (June 2009) (Learn how and when to remove this template
-
unctionals show that sometimes simply tethering two useful molecules together unlocks some useful activity⏎There’s been a real surge in drug discovery involving a class of molecules that (for a long time, at least) were considered at best an academic curiosity.
-
) the right to amend that record if it is inaccurate, irrelevant, untimely, or incomplete, and (3) the right to sue the government for violations of the statute including permitting others
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: the potential of a computer) or a computer) or a computer) or acomputer) or acomputer) ora computer) ora computer) ora computer) ora computer) ora computer) ora computer) |a
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
ilohertz (kHz) passive integrated transponder (PIT) tag. Twenty subadults of each species were implanted with short-term (3 month) acoustic telemetry tags to examine dispersal patterns immediately following release. Ten
-
you to eat more produce: The government has gotten on the bandwagon. According to the U.S. Department of Agriculture’s (USDA) and Department of Health and Human Services’ (HHS) Dietary
-
and the belief that we can recover from setbacks and learn from them.⏎Singaporean students have typically scored very well in the Organisation for Economic Co-operation and Development’s (OECD) Programme for International Student Assessment (PISA) in Reading, Mathematics
-
over a period of 15 years (1996 to 2011) with different levels of harvesting intensity, including selective harvesting at low (13.0% removal of growing stock volume), medium (29.
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
- L5:243 suppresses
- L9:339 excites
- L3:10 excites
- L8:189 suppresses
- L9:2300 excites
- L4:769 suppresses
- L9:2249 suppresses
- L6:854 excites comma after an introductory word like then or when
- L9:1371 excites
- L0:86 excites
- L7:288 excites
- L4:2293 excites
- L5:937 suppresses
- L4:2958 excites
- L2:353 excites
- L5:275 excites
- L9:284 suppresses
- L6:144 excites
- L3:2051 excites
- L4:64 excites
- L4:2309 excites
- L1:102 excites
- L0:1860 weakly excites
- L9:935 weakly suppresses
Feeds later layers
- L14:90 strongly suppresses
- L15:770 strongly suppresses
- L16:455 strongly suppresses
- L16:264 suppresses opening parenthesis in citations and asides
- L16:1767 suppresses
- L18:1365 suppresses
- L18:1140 suppresses
- L18:1247 suppresses
- L16:183 excites
- L19:1099 suppresses
- L18:357 suppresses
- L13:1721 excites
- L16:1253 suppresses
- L17:238 suppresses
- L11:242 suppresses
- L16:332 suppresses plural '(s)' in parentheses
- L16:2853 suppresses
- L19:1762 suppresses
- L18:721 suppresses opening parenthesis after a word
- L14:220 suppresses
- L15:277 suppresses
- L19:1864 suppresses
- L16:1494 suppresses
- L15:1193 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
- ovy
- (Parameter
- `'
- ammen
- GORITHM
- owitz
- ��
- `.
- hyper
- cyl
Suppresses
- int
- 受
- indo
- 礼
- ida
- ()},
- *);⏎
- }",
- frog
- ”。⏎⏎